Anthropic
Last reviewed
May 30, 2026
Sources
111 citations
Review status
Source-backed
Revision
v9 ยท 14,446 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 30, 2026
Sources
111 citations
Review status
Source-backed
Revision
v9 ยท 14,446 words
Add missing citations, update stale details, or suggest a clearer explanation.
| Anthropic PBC | |
|---|---|
 | |
| Type | Private Public-benefit corporation |
| Industry | Artificial intelligence |
| Founded | 2021 |
| Founders | Dario Amodei (CEO) Daniela Amodei (President) Tom Brown Chris Olah Sam McCandlish Jack Clark Jared Kaplan Benjamin Mann |
| Headquarters | San Francisco, California, United States |
| Key people | Dario Amodei (CEO) Daniela Amodei (President) Mike Krieger (Co-lead of Labs) Rahul Patil (CTO) Ami Vora (Head of Product) Krishna Rao (CFO) Paul Smith (CCO) Jason Clinton (CISO) Jan Leike (Co-lead of Alignment Science) |
| Products | Claude (AI assistant, current flagship Claude Opus 4.8) Claude API Claude Code Claude Cowork Claude Marketplace Claude Mythos Preview (restricted) |
| Revenue | ~$47 billion run-rate (May 2026)[108] |
| Employees | ~3,000 (2026)[2] |
| Website | anthropic.com |
Anthropic PBC is an American artificial intelligence (AI) safety and research company founded in 2021. The company develops and deploys large language models (LLMs), most notably the Claude family of AI assistants, with a stated focus on AI safety and alignment. Its goal is to ensure that advanced AI systems are reliable, interpretable, and steerable.[3] As of February 2026, Anthropic was valued at $380 billion following a $30 billion Series G funding round. On May 28, 2026, the company closed a $65 billion Series H round at a $965 billion post-money valuation, becoming the most valuable AI startup, and disclosed that its run-rate revenue had crossed $47 billion earlier that month.[4][79][108]
Anthropic was founded in 2021 by eight former employees of OpenAI, including siblings Dario Amodei and Daniela Amodei. Dario, who served as OpenAI's Vice President of Research, became Anthropic's CEO, while Daniela, formerly VP of Safety & Policy at OpenAI, became President.[5] The founders left OpenAI due to directional differences, particularly concerns about AI safety and the pace of AI development.[3]
The company was initially incorporated as a Delaware public benefit corporation (PBC), a structure that enables directors to balance stockholders' financial interests with its public benefit purpose of "the responsible development and maintenance of advanced AI for the long-term benefit of humanity."[6]
In May 2021, Anthropic raised $124 million in Series A funding led by Jaan Tallinn, co-founder of Skype.[7] In April 2022, the company raised $580 million in Series B funding, including $500 million from FTX prior to its collapse.[8]
The company spent 2022 developing the first version of Claude but chose not to release it publicly, citing the need for internal safety testing and a desire to avoid initiating a potentially hazardous race in AI capabilities development.[9]
In March 2023, Anthropic publicly launched Claude, its first AI assistant, marking the company's entry into the competitive AI assistant market.[10] In July 2023, the company was named to the White House's voluntary AI safety commitments alongside other leading labs like OpenAI, Google, and Meta.[11]
Throughout 2023, the company secured major investments and partnerships:
May 2023: Raised $450 million in Series C funding led by Spark Capital[12]
September 2023: Amazon announced an initial investment of up to $4 billion, making AWS Anthropic's primary cloud provider[13]
October 2023: Google invested $2 billion in the company[14]
In March 2024, Anthropic released the Claude 3 family of models (Haiku, Sonnet, and Opus), with Opus outperforming OpenAI's GPT-4 and Google's Gemini Ultra on various benchmarks.[15] June 2024 saw the release of Claude 3.5 Sonnet, which demonstrated significant improvements over Claude 3 Opus despite being a smaller model, alongside the introduction of the Artifacts feature for real-time code and content generation.[16] In October 2024, Anthropic released an upgraded Claude 3.5 Sonnet alongside Claude 3.5 Haiku, the fastest model in its lineup, which scored 40.6% on SWE-bench Verified and featured a 200,000-token context window.[17]
Also in November 2024, Anthropic announced the Model Context Protocol (MCP), an open standard for connecting AI assistants to external data systems and tools. MCP was designed to solve the "N-by-M" integration problem, where developers previously had to build custom connectors for each data source. The protocol was quickly adopted by major AI providers including OpenAI and Google DeepMind, and by early 2026 had become the de facto standard for connecting agents to tools and data, with thousands of community-built MCP servers and SDKs in Python, TypeScript, C#, and Java.[18]
In November 2024, Amazon invested an additional $4 billion, bringing its total investment to $8 billion and establishing AWS as Anthropic's primary training partner.[19]
The year 2025 was transformative for Anthropic. In February, the company released Claude 3.7 Sonnet, billed as the first "hybrid reasoning model," capable of producing both near-instant responses and extended step-by-step thinking visible to the user. Claude 3.7 Sonnet achieved 70.3% on SWE-bench Verified and was released alongside an early preview of Claude Code, an agentic coding tool.[20]
In March 2025, Anthropic raised $3.5 billion in Series E funding at a $61.5 billion valuation, led by Lightspeed Venture Partners.[21]
On May 22, 2025, Anthropic introduced the Claude 4 generation: Claude Opus 4 and Claude Sonnet 4. Opus 4 led on SWE-bench at 72.5% and Terminal-bench at 43.2%, and could work continuously for several hours on complex, long-running agent workflows. Sonnet 4 matched Opus 4's SWE-bench score (72.7%) at a fraction of the cost. Both models introduced extended thinking with tool use, allowing Claude to alternate between reasoning and tool calls within a single turn. Anthropic also activated ASL-3 (Anthropic Safety Level 3) safeguards for these models.[22]
Claude Code was fully released in May 2025, and its growth was explosive. Within six months of launch it reached a $1 billion annualized run rate, a velocity that even ChatGPT did not match.[23]
In July 2025, Anthropic accepted a $200 million contract from the United States Department of Defense for military AI applications, though the contract included usage restrictions negotiated by Anthropic.[24] In August 2025, the company released Claude Opus 4.1.[25]
In September 2025, Anthropic completed Series F funding of $13 billion at a $183 billion valuation, co-led by ICONIQ Capital, Fidelity Management, and Lightspeed Venture Partners. The company also released Claude Sonnet 4.5, marketed as its strongest model for coding, agents, and computer use.[26] That same month, Anthropic agreed to pay $1.5 billion to settle a class action copyright lawsuit (Bartz v. Anthropic) brought by authors who alleged the company had used millions of pirated books from shadow libraries to train its models. The settlement, preliminarily approved by Judge William Alsup, covered approximately 500,000 works at roughly $3,000 per book, making it the largest copyright settlement in U.S. history.[27]
In October 2025, Anthropic hired former Stripe CTO Rahul Patil as its new Chief Technology Officer. As part of this change, co-founder Sam McCandlish moved to a new role as Chief Architect, and the company restructured its core technical group to bring its product-engineering team into closer contact with the infrastructure and inference teams. Mike Krieger stepped away from his Chief Product Officer role to co-lead Anthropic's Labs team, with Ami Vora taking over as Head of Product.[63]
The company's annualized run-rate revenue grew from $1 billion at the start of 2025 to over $9 billion by the end of the year.[28]
On November 24, 2025, Anthropic launched Claude Opus 4.5, the first AI model to break 80% on SWE-bench Verified (scoring 80.9%). The model came with a 67% price cut relative to the previous Opus tier, priced at $5 per million input tokens and $25 per million output tokens.[29]
In December 2025, Anthropic acquired Bun, the all-in-one JavaScript runtime created by Jarred Sumner, with the company emphasising that Bun would remain open source and MIT-licensed. The announcement noted that Claude Code had already reached a $1 billion annual run-rate within six months of launch and that Bun had become "essential infrastructure for AI-led software engineering."[80] Also in December 2025, Anthropic donated MCP to the newly established Agentic AI Foundation (AAIF), a directed fund under the Linux Foundation. The AAIF was co-founded by Anthropic, Block, and OpenAI, with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg. MCP joined two other founding projects: goose (by Block) and AGENTS.md (by OpenAI). In one year, MCP had grown to over 97 million monthly SDK downloads and 10,000 active servers, with first-class client support across major AI platforms including ChatGPT, Claude, Cursor, Gemini, Microsoft Copilot, and Visual Studio Code.[64]
Anthropic entered 2026 with continued momentum.
On January 12, 2026, Anthropic launched Claude Cowork as a research preview, a general-purpose AI agent built into the Claude Desktop app on macOS. Described as "Claude Code for the rest of your work," Cowork allows users to designate a folder on their computer where Claude can read, edit, and create files. Use cases demonstrated at launch included reorganizing downloads, converting receipt screenshots into expense spreadsheets, and producing first drafts from scattered notes. The tool implements sub-agent coordination for parallelizable tasks, spawning multiple Claude instances that execute concurrently. At launch, Cowork was available only to Max subscribers ($100 to $200 per month).[70]
On February 5, 2026, the company released Claude Opus 4.6, featuring a 1 million token context window in beta (the first Opus-class model to support this), support for up to 128k output tokens (double the previous limit), and a new "agent teams" capability. Agent teams allow multiple Claude Code agents to work in parallel from a single orchestrator, each with its own full context window, communicating through what Anthropic calls a "Mailbox Protocol." In a demonstration, 16 parallel agents wrote a 100,000-line C compiler (in Rust) in two weeks that could compile the Linux 6.9 kernel, QEMU, FFmpeg, SQLite, PostgreSQL, and Redis with a 99% pass rate on the GCC test suite.[30] Opus 4.6 also introduced adaptive thinking as its recommended reasoning mode, where Claude dynamically decides when and how much to think.[31]
On February 12, 2026, Anthropic closed a $30 billion Series G funding round at a $380 billion post-money valuation. The round was led by GIC (Singapore's sovereign wealth fund) and Coatue, with D. E. Shaw Ventures, Founders Fund, and Abu Dhabi's MGX co-leading. Other participants included Sequoia Capital, BlackRock, Blackstone, Temasek, Goldman Sachs, JPMorgan Chase, and Fidelity. This was the second-largest private financing round in tech history, trailing only OpenAI's $40 billion raise.[4]
On February 17, 2026, Anthropic released Claude Sonnet 4.6, which achieved 79.6% on SWE-bench Verified and 72.5% on OSWorld. In developer testing, Sonnet 4.6 was preferred over the previous flagship Opus 4.5 59% of the time for coding tasks, at one-fifth the price ($3/$15 per million tokens). It became the default model for Free and Pro plan users on claude.ai.[32]
Also on February 17, 2026, Anthropic announced a strategic collaboration with Infosys to develop enterprise AI solutions across telecommunications, financial services, manufacturing, and software development. The partnership integrates Anthropic's Claude models and Claude Code with Infosys Topaz AI offerings, with a focus on building agentic AI systems using the Claude Agent SDK. The collaboration began with a dedicated Anthropic Center of Excellence for the telecommunications sector. Anthropic also formally opened its Bengaluru office, its second in Asia after Tokyo, led by country Managing Director Irina Ghose. India had become Claude's second-largest market worldwide, accounting for roughly 6% of global usage.[71]
On February 20, 2026, Anthropic launched Claude Code Security as a limited research preview for Enterprise and Team customers. Claude Code Security scans codebases for security vulnerabilities by reasoning about code the way a human security researcher would: understanding how components interact, tracing data flows across files, and catching complex multi-component vulnerability patterns that static analysis tools miss. Each identified vulnerability goes through a multi-stage verification process to filter out false positives and is assigned a severity rating. Nothing is applied without human approval. Using Opus 4.6, the team discovered over 500 high-severity vulnerabilities in widely used open-source libraries such as Ghostscript, OpenSC, and CGIF, many of which had gone undetected for decades despite years of expert review. Open-source maintainers received expedited access to the tool.[72]
On February 24, 2026, Anthropic published Version 3.0 of its Responsible Scaling Policy (RSP), introducing Frontier Safety Roadmaps and Risk Reports while restructuring the framework around two sets of mitigations: unilateral commitments Anthropic will pursue regardless of industry behaviour, and a more ambitious capabilities-to-mitigations map that would require multi-firm cooperation. The update explicitly acknowledged that scaling protections to ASL-4 and beyond cannot be done by any single firm in isolation, particularly against state-level adversaries, and reframed certain previously announced safeguards as goals contingent on shared effort. ASL-3 remained the operative posture for Anthropic's frontier models, with ASL-4 capability thresholds defined but not yet activated.[56]
Also on February 24, 2026, Anthropic publicly accused three Chinese AI companies, DeepSeek, Moonshot AI, and MiniMax, of conducting "industrial-scale" distillation attacks on Claude. According to Anthropic's investigation, the three firms created over 24,000 fraudulent accounts and generated more than 16 million exchanges with Claude to extract its capabilities. MiniMax accounted for the bulk of the traffic with over 13 million exchanges targeting agentic coding and tool use. Moonshot AI generated more than 3.4 million exchanges aimed at agentic reasoning. DeepSeek conducted over 150,000 exchanges focused on foundational logic and alignment, with prompts designed to reveal chain-of-thought training data. Anthropic warned that models built through illicit distillation are unlikely to retain safety protections.[65]
On February 25, 2026, Anthropic acquired Vercept, a Seattle-based AI startup specializing in computer-use agents. Vercept had developed Vy, a computer-use agent capable of operating a remote Apple MacBook in the cloud. Co-founders Kiana Ehsani, Luca Weihs, and Ross Girshick joined Anthropic. The acquisition was intended to advance Claude's computer use capabilities, enabling it to complete multi-step tasks inside live applications.[66]
The Enterprise Analytics API was also introduced in February 2026, giving Enterprise plan customers programmatic access to usage and engagement data. The API provides five endpoints covering per-user activity metrics, organization-wide activity summaries, chat project usage, skill usage, and Claude Code metrics (including commits, pull requests, and lines of code). Data is aggregated per organization, per day, with a default rate limit of 60 requests per minute.[73]
The Pentagon relationship deteriorated sharply in early 2026. In January, Defense Secretary Pete Hegseth released a new AI strategy calling for contracts with AI companies to eliminate company-specific guardrails. Anthropic refused to accept language permitting unrestricted military use of its models, with Dario Amodei insisting on retaining prohibitions against mass domestic surveillance and fully autonomous weapons systems.[33]
In March 2026, Anthropic made its memory feature available to all users, including those on the free plan. The feature, previously limited to paid subscribers since October 2025, allows Claude to remember user preferences and context across conversations. Anthropic also introduced a memory import tool that lets users transfer their preferences from competing chatbots like ChatGPT and Gemini.[34]
On March 3, 2026, the Department of Defense formally designated Anthropic a supply chain risk to national security, the first such designation ever applied to an American company. On March 9, Anthropic filed lawsuits in two federal courts challenging the designation, arguing that it punished the company for exercising protected speech on AI policy. Nearly 150 retired federal and state judges subsequently filed an amicus brief supporting Anthropic, alongside statements of support from Microsoft, Google engineers, and employees at competing AI companies. The financial impact is significant: Anthropic estimated the designation could cost hundreds of millions or even billions of dollars in lost government revenue for 2026. OpenAI subsequently signed a contract with the Defense Department to fill the gap.[33]
On March 6, 2026, Anthropic launched the Claude Marketplace, a curated enterprise platform where businesses can purchase Claude-powered tools from third-party partners. Enterprises with existing Anthropic spend commitments can redirect part of their budget toward partner applications, consolidating procurement under a single invoice. Launch partners included Snowflake, GitLab, Harvey, Replit, Rogo, and Lovable. Notably, Anthropic does not take a revenue cut from Marketplace purchases, distinguishing it from cloud marketplace models operated by Amazon Web Services and Microsoft Azure.[74]
On March 9, 2026, Anthropic released Code Review for Claude Code, a multi-agent code review system that automatically analyzes pull requests for bugs, logic errors, and security vulnerabilities. When a pull request is opened, Code Review dispatches a team of agents that work in parallel. Findings go through a verification step to filter out false positives, and confirmed issues are ranked by severity. Results appear directly on the pull request as a single overview comment with inline comments for specific bugs. Agents do not approve pull requests; developers always retain final authority. In Anthropic's internal testing, the percentage of pull requests receiving substantive review comments rose from 16% to 54%, and on large pull requests (over 1,000 lines changed), 84% received findings averaging 7.5 issues. Code Review is available as a research preview for Team and Enterprise customers, with token-based pricing averaging $15 to $25 per review. Reviews typically take around 20 minutes per pull request.[75]
On March 10, 2026, Anthropic announced it would open an office in Sydney, its fourth in the Asia-Pacific region after Tokyo, Bengaluru, and Seoul. Australia ranked fourth globally in Claude.ai usage, with New Zealand ranking eighth. Existing Australian customers include Canva, Quantium, and Commonwealth Bank of Australia.[67]
On March 11, 2026, Anthropic launched the Anthropic Institute, a new research arm led by co-founder Jack Clark (in a new role as Head of Public Benefit). The Institute brings together three existing teams: the Frontier Red Team, which stress-tests AI systems; Societal Impacts, which studies how AI is used in the real world; and Economic Research, which tracks AI's effects on jobs and the broader economy. The Institute also incubates new research efforts, including forecasting AI progress and studying how advanced AI will interact with the legal system. Notable hires for the Institute include Matt Botvinick (formerly of Google DeepMind and Princeton), Anton Korinek (on leave from the University of Virginia), and Zoe Hitzig (previously at OpenAI). Anthropic simultaneously expanded its Public Policy organization under Sarah Heck as Head of Public Policy.[68]
On March 12, 2026, Anthropic launched the Claude Partner Network with a $100 million investment. Anchor partners include Accenture, Deloitte, Cognizant, and Infosys. The program offers training, dedicated technical support, joint go-to-market investment, and the first Claude technical certification (Claude Certified Architect, Foundations). Anthropic is scaling its partner-facing headcount fivefold, adding Applied AI engineers, technical architects, and localized go-to-market support across international markets.[69]
On March 17, 2026, Anthropic announced Claude Dispatch, a feature inside Claude Cowork that creates a single persistent conversation between the Claude mobile app on a phone and the Claude Desktop app on a computer. Users can assign tasks from their phone (such as summarizing recent work, pulling together a brief, comparing materials across folders, or producing a report) and return to completed work on their desktop. Execution occurs in a local sandboxed environment where files never leave the user's computer. Dispatch supports over 38 connectors including Notion, Gmail, Slack, Google Calendar, Google Drive, Dropbox, GitHub, Figma, Trello, and Asana. Max plan subscribers received immediate access, with Pro plan users gaining access within a few days.[76]
On March 20, 2026, Anthropic shipped Claude Code Channels as a research preview, enabling developers to connect a running Claude Code session to Telegram or Discord. Messages sent through the chat app are picked up by the local Claude Code session, which executes the work and replies through the same platform. The feature uses the Model Context Protocol (MCP): a Channel is an MCP server that declares the claude/channel capability and actively pushes events into a running session. Channels can be one-way (forwarding alerts or CI notifications) or two-way. The security model includes an allowlist-based plugin system (only Anthropic-approved plugins during the preview), pairing-code authentication that locks the bot to a specific user ID, and prompt injection threat modeling. The feature requires Claude Code version 2.1.80 or later and supports Telegram and Discord as of launch.[77]
On March 23, 2026, Anthropic launched Computer Use for Mac as a research preview, giving Claude the ability to control a Mac computer directly. Claude can move the mouse, use the keyboard, open applications, navigate browsers, fill in spreadsheets, and complete multi-step tasks autonomously. When a supported connector is available (such as Google Workspace or Slack), Claude prioritizes that integration, but it falls back to screen-based control when no connector exists. Claude requests permission before accessing new applications, and users can halt operations at any point. The feature is available to Pro and Max subscribers on macOS and integrates with Dispatch, allowing users to assign computer-use tasks from their phone. Windows and Linux support has not yet been announced.[78]
By March 2026, Anthropic's annualized run-rate revenue had surged to $19 billion, more than doubling from $9 billion at the end of 2025. Claude Code alone generated over $2.5 billion in annualized revenue. CEO Dario Amodei confirmed at a Morgan Stanley TMT conference that the company added $6 billion in run-rate revenue during February 2026 alone. Anthropic was capturing over 73% of all spending among companies buying AI tools for the first time.[1]
Anthropic has also been preparing for a potential IPO, reportedly hiring the law firm Wilson Sonsini and holding preliminary talks with investment banks, with reporting from The Information citing internal discussions targeting an October 2026 listing window. Anthropic itself has not publicly committed to a timeline.[35]
April 2026 was an unusually dense month of product, partnership, and capital announcements.
On April 3, 2026, multiple reports indicated that Anthropic had acquired Coefficient Bio, an eight-month-old stealth startup using AI to make drug discovery and biological research more efficient, in a stock deal valued at just over $400 million. Coefficient Bio's founders Samuel Stanton and Nathan C. Frey had previously worked on computational drug discovery at Genentech's Prescient Design. The roughly 10-person team joined Anthropic's health and life sciences group, complementing the Claude for Life Sciences product line.[81]
On April 7, 2026, Anthropic announced Project Glasswing, an industry-wide cybersecurity initiative anchored by 12 founding partners: AWS, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. The coalition operates around Claude Mythos Preview, an unreleased frontier model that Anthropic described as a "step change" in cyber capability. Mythos Preview had identified thousands of zero-day vulnerabilities across major operating systems and web browsers during pre-launch testing, including flaws in OpenBSD code that had survived 27 years of human review, FFmpeg code 16 years old, and parts of the Linux kernel. On the CyberGym benchmark, Mythos Preview scored 83.1%, compared with 66.6% for Opus 4.6. Anthropic committed up to $100 million in Mythos Preview usage credits to Glasswing partners and approved organizations maintaining critical software infrastructure, $2.5 million in donations to Alpha-Omega and OpenSSF through the Linux Foundation, and $1.5 million to the Apache Software Foundation. Anthropic stated it does not plan to release Mythos generally. Following the research preview period, the model is priced at $25 per million input tokens and $125 per million output tokens, accessible via the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.[82][83]
On April 9, 2026, Anthropic shipped a triple announcement: Claude Cowork moved from research preview to general availability on macOS and Windows for all paying subscribers, gaining six enterprise features (role-based access controls, group spend limits, expanded usage analytics, OpenTelemetry support, a Zoom MCP connector, and per-tool connector controls); Claude Managed Agents entered public beta as a suite of composable APIs and a fully managed cloud runtime for deploying agents at enterprise scale at $0.08 per session-hour on top of standard token pricing; and Claude Code received a major update with policy controls and an Amazon Bedrock setup wizard. Early Managed Agents customers included Notion, Rakuten, and Asana.[84]
On April 14, 2026, Anthropic opened the research preview of Claude Code Routines, a cloud-hosted automation tier that allows developers to save a Claude Code configuration (a prompt, one or more repositories, and a set of connectors) and run it on a schedule, on an API call, or in response to a GitHub event. Routines survive restarts, terminal closures, and overnight runs, distinguishing them from the desktop-bound Scheduled Tasks shipped in February. Pro users can run five routines per day, Max users 15, and Team and Enterprise users 25.[85]
On April 16, 2026, Anthropic released Claude Opus 4.7, its new flagship model. Opus 4.7 substantially improved on Opus 4.6 in advanced software engineering, reaching 64.3% on SWE-bench Pro (versus 53.4% for Opus 4.6 and 57.7% for GPT-5.4) and 78.0% on OSWorld. Vision was the largest single upgrade: the model accepts images up to 2,576 pixels on the long edge (about 3.75 megapixels and roughly 3.3x the prior ceiling), with image coordinates mapping 1:1 to actual pixels. Internal vision-accuracy benchmarks rose from 54.5% to 98.5%. The release introduced a new "xhigh" effort level between high and max, task budgets that expose a countdown to the model during agentic loops, and file-system-based memory for multi-session work. Pricing remained $5 per million input tokens and $25 per million output tokens. Opus 4.7 was made available through the Claude API as claude-opus-4-7, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Anthropic noted that the model carries a similar safety profile to Opus 4.6 but added automatic detection of prohibited or high-risk cybersecurity requests, paired with a new Cyber Verification Program for authorized researchers.[86]
On April 17, 2026, Anthropic Labs launched Claude Design in research preview for Pro, Max, Team, and Enterprise subscribers, a product for creating prototypes, slides, one-pagers, and other visual work through conversation with Opus 4.7. Designs can be edited inline or via custom sliders, automatically conform to a team's design system when one is provided, and export to Canva, PDF, PPTX, or standalone HTML. Figma's publicly traded stock declined approximately 7% the day of the announcement.[87]
On April 24, 2026, Anthropic announced a wide-ranging strategic collaboration with NEC Corporation under which Claude will be made available to approximately 30,000 NEC Group employees worldwide. NEC became Anthropic's first Japan-based global partner. Components include co-developing AI products for the Japanese market starting with finance, manufacturing, and local-government use cases; integrating Claude into NEC's Security Operations Center; embedding Opus 4.7 and Claude Code into NEC's BluStellar Scenario consulting offerings; and standing up an NEC Center of Excellence trained by Anthropic.[88]
Also in late April, Google and Amazon dramatically expanded their financial commitments to Anthropic. Google agreed to invest up to $40 billion in cash and compute, with $10 billion initial at a $350 billion valuation and an additional $30 billion contingent on performance milestones. Amazon committed an additional $5 billion at a $350 billion valuation, with plans to invest up to $20 billion more over time. The new deals included an agreement for Anthropic to secure up to 5 gigawatts of compute capacity from AWS. Together with prior commitments, Amazon's and Google's stakes had grown to a level at which Fortune reported that roughly half of each company's headline AI-related earnings during the period derived from mark-to-market gains on the Anthropic position rather than from operating profit.[89][90]
On April 27, 2026, Anthropic formally opened its Sydney office and named Theo Hourmouzis, previously Senior Vice President for Australia, New Zealand and ASEAN at Snowflake, as General Manager for Australia and New Zealand. The Sydney opening was accompanied by new partnerships with Canva (integrating Canva's Design Engine and Visual Suite into Claude Design), Xero (a multi-year deal connecting Xero's financial data with Claude.ai), and four Australian universities.[91]
On April 28, 2026, Anthropic launched Claude for Creative Work along with nine new connectors for professional creative software, available across all Claude plans. The Adobe connector exposes more than 50 tools across Photoshop, Lightroom, Illustrator, Firefly, Premiere, Express, InDesign, and Adobe Stock via MCP. Additional connectors covered Autodesk Fusion (text-to-3D-modeling), Blender (natural-language access to Blender's Python API), Ableton, and Splice. The launch was paired with educational partnerships with the Rhode Island School of Design, Ringling College of Art and Design, and Goldsmiths, University of London.[92]
On May 4, 2026, Anthropic, Blackstone, Hellman & Friedman, and Goldman Sachs announced the formation of a new AI-native enterprise services company designed to bring Claude into the core operations of mid-sized companies, with an initial focus on community banks, manufacturers, regional health systems, and private-equity portfolio companies. The new firm is a standalone entity with Anthropic engineering and partnership resources embedded directly, and is also backed by General Atlantic, Leonard Green, Apollo Global Management, GIC, and Sequoia Capital. CNBC reported the venture as a $1.5 billion firm. The company joins the Claude Partner Network alongside Accenture, Deloitte, Cognizant, Infosys, and PwC.[93][94]
On May 5, 2026, Anthropic held an invite-only "Claude for Financial Services" briefing in New York where it unveiled ten pre-built AI agent templates for financial institutions: pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, valuation reviewer, general ledger reconciler, month-end closer, statement auditor, and KYC screener. The launch was paired with new connectors and MCP apps from Dun & Bradstreet, Fiscal AI, Financial Modeling Prep, Guidepoint, IBISWorld, SS&C Intralinks, Third Bridge, Verisk, and a Moody's MCP app delivering proprietary credit ratings and data on more than 600 million public and private companies. The event was headlined by a joint appearance of Dario Amodei and JPMorgan Chase Chairman and CEO Jamie Dimon, who demonstrated Claude Code building a Treasury bid-ask spread dashboard in 20 minutes. Microsoft 365 Copilot's integration with Claude was extended through new add-ins to Excel, PowerPoint, Word, and Outlook.[95]
On May 6, 2026, Anthropic and SpaceX announced a compute deal that gave Anthropic access to more than 300 megawatts of new capacity (over 220,000 NVIDIA GPUs) at SpaceX's Colossus One data center within the month. Anthropic concurrently doubled the five-hour Claude Code rate limit for Pro, Max, Team, and seat-based Enterprise subscribers, eliminated peak-hour throttling for Pro and Max accounts, and raised API rate limits for Tier 1 customers by 1,500% on input tokens per minute and 900% on output tokens per minute. Free-tier accounts were excluded from the capacity increases.[96]
On May 7, 2026, Anthropic shipped Claude as native add-ins inside Excel, PowerPoint, and Word at general availability, with an Outlook add-in in public beta. The four integrations share a single conversation thread per Microsoft 365 user, allowing context (such as financial model assumptions) to propagate across applications. The add-ins are included in all paid Claude plans at no extra charge and install via the Microsoft Marketplace.[97]
On May 12, 2026, Anthropic launched Claude for Legal with 12 practice-area plugins (Commercial Legal, Corporate Legal, Employment Legal, Privacy Legal, Product Legal, Regulatory Legal, AI Governance Legal, IP Legal, Litigation Legal, and others) and more than 20 MCP connectors for legal-specific systems including DocuSign, Ironclad, and Datasite. Anthropic noted that legal professionals had become the most engaged Claude Cowork users of any knowledge-work function.[98]
Also on May 12, 2026, Bloomberg reported that Anthropic was in early talks with investors to raise at least $30 billion in fresh financing at a valuation of more than $900 billion, with a non-binding round potentially closing within weeks. The reporting cautioned that no term sheet had been signed and Anthropic had not commented publicly on the talks.[79]
On May 13, 2026, Anthropic launched Claude for Small Business, a product running on Claude Cowork that connects Claude to QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365 with built-in workflows across finance, operations, sales, marketing, HR, and customer service. The launch included 15 ready-made skills covering tasks such as payroll planning, bookkeeping, and employee onboarding. The same day, Anthropic announced that effective June 15, 2026, programmatic Claude usage (Claude Agent SDK, the claude -p command, GitHub Actions, and third-party agent tools authenticating through subscriptions) would move to a separate metered credit pool priced at standard API list rates, with allocations mirroring subscription prices ($20 for Pro, $100 for Max 5x, $200 for Max 20x). Interactive use of Claude Code, Cowork, and chat remained on subscription limits. The change generated substantial developer backlash on X, with Anthropic's clarification tweet receiving a Community Note.[99][100]
On May 14, 2026, Anthropic kicked off the Claude for Small Business AI Fluency Tour in Chicago, a series of free half-day workshops for 100 local small-business leaders per stop in partnership with Tenex.co. Spring 2026 cities include Chicago, Tulsa, Dallas, Hamilton Township, Baton Rouge, Birmingham, Salt Lake City, Baltimore, San Jose, and Indianapolis. Anthropic also released "AI Fluency for Small Business," a free online course produced in partnership with PayPal, and partnered with Workday and the Local Initiatives Support Corporation on the Workday Foundation Solopreneurship Accelerator Program for an initial cohort of 15 entrepreneurs.[101]
Also on May 14, 2026, Anthropic and the Gates Foundation announced a $200 million, four-year partnership combining grant funding, Claude usage credits, and technical support for programs in global health, life sciences, education, and economic mobility. The largest single pillar focuses on accelerating vaccine and therapy development in low- and middle-income countries, with additional work on AI-powered tutoring for K-12 students in the United States, foundational literacy and numeracy programs in sub-Saharan Africa and India, and agricultural productivity tools for smallholder farmers. The partners agreed to release agriculture-specific Claude improvements, datasets for African languages, and benchmarks as public goods.[102]
On May 14, 2026, Anthropic and PwC also announced a major expansion of their alliance focused on agentic technology build, AI-native deal-making, and reinvention of the enterprise function. PwC committed to training and certifying 30,000 professionals on Claude, rolling out Claude Code and Cowork starting with U.S. teams, and standing up a new Finance Business Group built around Claude, Claude Cowork, and Claude Code. PwC reported delivery improvements of up to 70% in production deployments with Claude.[103]
Anthropic was also reportedly in advanced talks in May 2026 to acquire Stainless, a developer-tools startup whose AI-powered compiler generates production-ready SDKs (Python, TypeScript, Kotlin, Go, Java, Ruby) and MCP servers from OpenAPI specs, for at least $300 million. Stainless's libraries are used by, among others, OpenAI, Google, and Meta; The Information reported the deal as not yet final.[104]
On May 8, 2026, Anthropic signed a seven-year, $1.8 billion cloud computing agreement with Akamai Technologies to expand the infrastructure serving Claude, adding to capacity commitments already in place with AWS, Google, and SpaceX. Akamai initially described the customer only as a leading frontier model provider before subsequent reporting confirmed Anthropic.[111]
On May 28, 2026, Anthropic closed a $65 billion Series H funding round at a $965 billion post-money valuation, the largest venture round in its history and one that lifted it above OpenAI (last valued at $852 billion) as the most valuable AI startup. The round was co-led by Altimeter Capital, Dragoneer, Greenoaks, Sequoia Capital, Capital Group, Coatue, D1 Capital Partners, GIC, ICONIQ, and XN, with strategic infrastructure partners Micron, Samsung, and SK hynix also participating. Roughly $15 billion of the total comprised previously committed investments from hyperscalers, including $5 billion from Amazon announced in April. Anthropic disclosed that its run-rate revenue had crossed $47 billion earlier in May, more than doubling from $19 billion in March. Press reporting characterized the round as likely Anthropic's last private financing ahead of a widely anticipated IPO, with The New York Times reporting a target debut in the fall of 2026, though Anthropic did not commit publicly to a timeline.[108][109]
Also on May 28, 2026, Anthropic released Claude Opus 4.8, its new flagship model. Opus 4.8 improved on Claude Opus 4.7 across benchmarks, raising its agentic coding (SWE-bench Pro) score from 64.3% to 69.2% and its agentic computer use (OSWorld) score from 82.8% to 83.4%, while making notable gains in honesty and the ability to work independently for longer. Pricing was unchanged from Opus 4.7 at $5 per million input tokens and $25 per million output tokens, with a faster mode (about 2.5x speed) priced at $10 and $50 per million tokens that Anthropic said was three times cheaper than the equivalent mode on previous models. The release added a "dynamic workflows" capability in Claude Code that lets the model plan a task and then run hundreds of parallel subagents within a single session, and user-selectable effort levels (high, extra, and max) on claude.ai and Claude Cowork. Anthropic reported that Opus 4.8 showed substantially lower rates of misaligned behavior than Opus 4.7. The model was made available through the Claude API as claude-opus-4-8, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.[110]
Claude is Anthropic's family of large language models designed to be helpful, harmless, and honest. The models are generative pre-trained transformers that have been fine-tuned using Constitutional AI and reinforcement learning from human feedback (RLHF). The current flagship model is Claude Opus 4.8, released on May 28, 2026.[36][110]
| Generation | Models | Release Date | Key Features |
|---|---|---|---|
| Claude 1 | Claude 1.0, 1.3, Claude Instant | March 2023 | Initial release; Claude Instant as lightweight version |
| Claude 2 | Claude 2.0, 2.1 | July 2023 | 100,000 token context window, public web interface at Claude.ai |
| Claude 3 | Haiku, Sonnet, Opus | March 2024 | 200,000 token context, vision capabilities, multimodal support |
| Claude 3.5 | Sonnet, Haiku | June-October 2024 | Computer use capability, improved coding, Artifacts feature |
| Claude 3.7 | Sonnet | February 2025 | Hybrid reasoning model, extended thinking, Claude Code early preview |
| Claude 4 | Sonnet 4, Opus 4 | May 2025 | Extended thinking with tool use, ASL-3 safeguards, MCP connector |
| Claude 4.1 | Opus 4.1 | August 2025 | Performance improvements |
| Claude 4.5 | Sonnet 4.5, Opus 4.5 | September-November 2025 | Opus 4.5 first to break 80% SWE-bench; 67% price cut |
| Claude 4.6 | Opus 4.6, Sonnet 4.6 | February 2026 | 1M token context (beta), agent teams, adaptive thinking, 128k output |
| Claude 4.7 | Opus 4.7 | April 2026 | 64.3% SWE-bench Pro; 2,576-pixel vision; xhigh effort; task budgets[86] |
| Claude 4.8 | Opus 4.8 | May 2026 | 69.2% SWE-bench Pro; dynamic workflows; cheaper fast mode; improved alignment[110] |
| Mythos | Mythos Preview (restricted) | April 2026 | Unreleased; 83.1% CyberGym; Project Glasswing only[82] |
Constitutional AI (CAI) is Anthropic's proprietary methodology for training AI systems to be aligned with human values. The approach involves providing AI systems with a set of principles (a "constitution") that guide their behavior.[37]
The CAI process involves two main phases:
Supervised Learning Phase: The model generates self-critiques and revisions based on constitutional principles
Reinforcement Learning Phase: The model is trained using AI feedback (RLAIF, or Reinforcement Learning from AI Feedback) rather than solely human feedback[38]
The constitution draws from various sources including:
The UN Universal Declaration of Human Rights
Trust and safety best practices
Principles from other AI research labs
Platform guidelines from technology companies[39]
Anthropic extended the constitutional approach to safety classifiers in early 2025. The first generation of Constitutional Classifiers reduced jailbreak success rates from 86% to 4.4% during a red-teaming exercise where 183 participants spent over 3,000 hours attempting to break the system. No universal jailbreak was discovered. A second generation, Constitutional Classifiers++, reduced the compute overhead from 23.7% to roughly 1% by using a two-stage architecture: a lightweight probe screens all traffic using Claude's internal activations, escalating suspicious exchanges to a more powerful classifier.[40]
Anthropic conducts extensive research into interpretability, the field of understanding the internal workings of complex AI models. In 2024, the company used a compute-intensive technique called "dictionary learning" to identify millions of features (patterns corresponding to concepts) within the Claude 3 Sonnet model.[41] This research aims to better understand, monitor, and control model behavior to enhance safety.
Announced in November 2024, the Model Context Protocol is an open-source framework that standardizes how AI systems integrate with external tools, data sources, and systems. MCP defines a client-server architecture with three components: hosts (the AI application), clients (which maintain connections to servers), and servers (which provide context and tool access). The protocol has been adopted across the industry and is supported by SDKs in multiple programming languages.[18]
In December 2025, Anthropic donated MCP to the Agentic AI Foundation (AAIF) under the Linux Foundation, ensuring vendor-neutral governance under the same stewardship model that supports Kubernetes, PyTorch, and Node.js. By the time of the donation, MCP had grown to over 97 million monthly SDK downloads and 10,000 active servers, making it one of the fastest-growing open-source projects in AI history.[64]
Claude Mythos Preview is an unreleased frontier model introduced as part of Project Glasswing on April 7, 2026. Anthropic describes Mythos as "a general-purpose, unreleased frontier model" with a step-change in cybersecurity capability, scoring 83.1% on the CyberGym vulnerability reproduction benchmark (versus 66.6% for Opus 4.6). The model has been used by Anthropic's red team and approved Glasswing partners to identify thousands of zero-day vulnerabilities in critical open-source and commercial software, including issues in OpenBSD code dating back 27 years and Linux kernel components. Anthropic has stated it does not intend to make Mythos generally available; access is gated to Glasswing partners and approved organizations responsible for critical infrastructure, priced at $25 per million input tokens and $125 per million output tokens via the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.[82][83]
Anthropic operates as a public benefit corporation (PBC), legally requiring it to balance stockholder interests with its public benefit mission. The company must regularly report on how it promotes public benefits to its owners.[42]
Anthropic established a unique governance structure called the Long-Term Benefit Trust (LTBT), a purpose trust designed to ensure the company remains focused on "the responsible development and maintenance of advanced AI for the long-term benefit of humanity." The Trust holds Class T shares that allow it to elect directors to Anthropic's board.[43]
As of April 2025, the Trust members include:
Neil Buddy Shah
Kanika Bahl
Zach Robinson
Richard Fontaine[44]
Current board members include:
Dario Amodei (CEO and Co-Founder)
Daniela Amodei (President and Co-Founder)
Yasmin Razavi
Jay Kreps
Reed Hastings[45]
| Round | Date | Amount | Lead Investors | Valuation |
|---|---|---|---|---|
| Series A | May 2021 | $124M | Jaan Tallinn | Undisclosed |
| Series B | April 2022 | $580M | FTX, Google | Undisclosed |
| Series C | May 2023 | $450M | Spark Capital | $4.6B |
| Series D | September 2023 | $4B (Amazon) | Amazon | $18.5B |
| Google Investment | October 2023 | $2B | N/A | |
| Amazon Additional | November 2024 | $4B | Amazon | N/A |
| Series E | March 2025 | $3.5B | Lightspeed Venture Partners | $61.5B |
| Series F | September 2025 | $13B | ICONIQ, Fidelity, Lightspeed | $183B |
| Series G | February 2026 | $30B | GIC, Coatue | $380B (post-money) |
| Amazon expansion | April 2026 | $5B initial (up to $25B) | Amazon | $350B[89] |
| Google expansion | April 2026 | $10B initial (up to $40B) | $350B[90] | |
| Series H | May 2026 | $65B | Altimeter, Dragoneer, Greenoaks, Sequoia | $965B (post-money)[108] |
Total funding raised: approximately $57 billion in closed rounds through the Series G (February 2026), followed by a $65 billion Series H in May 2026 at a $965 billion post-money valuation, alongside April 2026 strategic commitments from Amazon and Google totaling up to $65 billion in cash and compute if all tranches and milestones are met.[46][108]
Amazon: $8 billion in prior investments plus a new $5 billion commitment in April 2026 (with up to $20 billion more contingent on the AWS compute deal). Reporting by Fortune in April 2026 noted that a significant share of Amazon's headline AI-related profits was attributable to mark-to-market gains on its Anthropic stake.[19][89]
Google: $2 billion initial investment plus an April 2026 commitment of up to $40 billion ($10 billion initial at a $350 billion valuation, with $30 billion contingent on performance milestones). Fortune's April 30, 2026 analysis attributed a similar share of Google's reported AI-related profit gains to the Anthropic stake.[14][90]
Microsoft: Up to $5 billion investment commitment, plus $30 billion Azure cloud services deal[47]
GIC (Singapore sovereign wealth fund): Co-led Series G; also backs the May 2026 enterprise services joint venture
Coatue: Co-led Series G
Lightspeed Venture Partners: Led multiple rounds
Fidelity Management: Co-led Series F
ICONIQ Capital: Co-led Series F
Other notable investors include Salesforce Ventures, Menlo Ventures, General Catalyst, BlackRock, Blackstone, Sequoia Capital, Temasek, Goldman Sachs, JPMorgan Chase, Qatar Investment Authority, D. E. Shaw Ventures, Founders Fund, and MGX (Abu Dhabi)[48]
Anthropic accelerated its M&A activity from late 2025 onwards. Disclosed or widely reported transactions include:
Bun (December 2025): JavaScript runtime created by Jarred Sumner; financial terms undisclosed. Bun remains open source and MIT-licensed.[80]
Vercept (February 2026): Seattle-based computer-use agent startup whose co-founders Kiana Ehsani, Luca Weihs, and Ross Girshick joined Anthropic.[66]
Coefficient Bio (April 2026, reported): Stealth biotech-AI startup; reported $400+ million stock deal. The roughly 10-person team joined the health and life sciences group.[81]
Stainless (May 2026, reported, in advanced talks): Developer-tools startup generating SDKs and MCP servers from OpenAPI specs; reported at no less than $300 million. Not finalized as of May 2026.[104]
The consumer-facing Claude assistant is available through:
Claude.ai: Web-based interface launched July 2023
Claude iOS/Android apps: Mobile applications
Claude Pro/Max: Premium subscription tiers with enhanced capabilities[49]
The Claude API provides programmatic access to Claude models for developers and businesses, supporting:
Text generation and analysis
Code generation and debugging
Vision capabilities (image analysis up to 2,576 pixels on the long edge with Opus 4.7)
Extended context windows (up to 1 million tokens with Opus 4.6, Sonnet 4.6, and Opus 4.7)[50][86]
Introduced in February 2026, the Enterprise Analytics API gives Enterprise plan customers programmatic access to organization-wide usage and engagement data. The API provides five endpoints: per-user activity metrics (conversation counts, messages sent, projects created, files uploaded, artifacts created, and Claude Code metrics like commits and pull requests), organization-wide activity summaries (daily, weekly, and monthly active user counts with seat utilization), chat project usage (conversation and user counts by project), and skill usage breakdowns for Claude and Claude Code sessions. Data is aggregated per organization, per day, available for up to the past 90 days (from January 1, 2026 onward), with a default rate limit of 60 requests per minute.[73]
Effective June 15, 2026, Anthropic moved programmatic Claude usage from subscription quotas to a metered "Agent SDK Credit pool" priced at standard API list rates. The change applies to the Claude Agent SDK, the claude -p command, Claude Code GitHub Actions, and third-party agent tools authenticating through Claude subscriptions (including OpenClaw, Conductor, Zed, and Jean). Monthly credit allowances mirror subscription pricing: $20 for Pro, $100 for Max 5x, $200 for Max 20x. Interactive use of Claude Code, Cowork, and chat remained on existing subscription limits. The announcement received significant developer pushback; Anthropic's clarification tweet was annotated with a Community Note on X.[99]
Launched as an early preview in February 2025 alongside Claude 3.7 Sonnet and fully released in May 2025, Claude Code is an agentic coding assistant that operates in the terminal. It can read and write code, run tests, use the command line, and coordinate multi-agent workflows with developer oversight. With the release of Opus 4.6 in February 2026, Claude Code gained the "agent teams" feature, allowing multiple agents to work on different parts of a codebase in parallel. Claude Code has generated over $2.5 billion in annualized run-rate revenue as of February 2026, with business subscriptions quadrupling since the start of the year.[23]
Released as a research preview on April 14, 2026, Routines are saved Claude Code configurations (prompt, repositories, and connectors) that run autonomously in the cloud. They can be triggered on a schedule, via API, or in response to GitHub events. Daily Routine quotas are 5 for Pro, 15 for Max, and 25 for Team and Enterprise users. Routines complement (and largely supersede) the desktop-bound Scheduled Tasks shipped in February 2026.[85]
Released on March 9, 2026, Code Review is a multi-agent system built into Claude Code that automatically analyzes pull requests for bugs, logic errors, and security vulnerabilities. When a pull request is opened, the system dispatches a team of agents that work in parallel, with findings verified to filter out false positives and ranked by severity. Results appear as a single overview comment on the pull request with inline comments for specific bugs. In Anthropic's internal testing, substantive review comments rose from 16% to 54% of pull requests, and 84% of large pull requests (over 1,000 lines changed) received findings. Code Review is available as a research preview for Team and Enterprise customers at an average cost of $15 to $25 per review.[75]
Launched on February 20, 2026 as a limited research preview, Claude Code Security scans codebases for security vulnerabilities by reasoning about code in the way a human security researcher would. Unlike traditional static analysis tools, it understands how components interact, traces data flows across files, and catches complex multi-component vulnerability patterns. Each finding goes through a multi-stage verification process. During development, the team used Opus 4.6 to discover over 500 high-severity vulnerabilities in widely used open-source libraries that had gone undetected for years. The tool is available to Enterprise and Team customers, with expedited access for open-source maintainers.[72]
Shipped on March 20, 2026 as a research preview, Claude Code Channels lets developers connect a running Claude Code session to Telegram or Discord. Messages sent through the chat app are picked up by the local session, which executes the requested work and replies through the same platform. The feature is built on the Model Context Protocol: a Channel is an MCP server that declares the claude/channel capability and pushes events into an active session. Channels can be one-way (forwarding alerts or CI notifications) or two-way (full conversational interaction). The security model uses an allowlist-based plugin system (only Anthropic-approved plugins during the preview) and pairing-code authentication. Claude Code version 2.1.80 or later is required.[77]
Introduced on January 12, 2026 as a research preview, Claude Cowork extends Claude's capabilities to everyday office tasks. Described as "Claude Code for the rest of your work," it is built into the Claude Desktop app and allows users to designate a folder on their computer where Claude can read, edit, and create files. The tool implements sub-agent coordination for parallelizable tasks, spawning multiple Claude instances that execute concurrently. Cowork launched for Max subscribers on macOS, with Pro subscribers and Team/Enterprise plans gaining access in subsequent weeks. By February 2026, Anthropic had launched a plugin marketplace and admin controls for enterprise deployments, along with connectors for Google Drive, Gmail, DocuSign, and other business tools. A Windows version followed in February 2026.[51][70] On April 9, 2026, Cowork reached general availability across macOS and Windows for all paying subscribers and gained six new enterprise features: role-based access controls, group spend limits, expanded usage analytics, OpenTelemetry support, a Zoom MCP connector, and per-tool connector controls.[84]
Announced on March 17, 2026, Claude Dispatch is a feature inside Claude Cowork that creates a single persistent conversation between the Claude mobile app on a phone and the Claude Desktop app on a computer. Users can assign tasks from their phone and return to completed work on their desktop. Execution occurs in a local sandboxed environment; files never leave the user's computer and no sensitive data is sent to Anthropic's servers. Dispatch supports over 38 connectors including Notion, Gmail, Slack, Google Calendar, Google Drive, Dropbox, GitHub, Figma, Trello, and Asana. Max plan subscribers received immediate access, with Pro plan users gaining access within days.[76]
Entering public beta on April 9, 2026, Claude Managed Agents is a suite of composable APIs and a fully managed cloud runtime that handles sandboxing, permission management, session continuity, and context management on Anthropic's infrastructure rather than on the developer's. The service is priced at $0.08 per session hour on top of standard API token pricing. Early customers include Notion, Rakuten, and Asana, who have used the platform to deploy agents across project management, HR, finance, and software development.[84]
Launched on March 6, 2026, the Claude Marketplace is a curated enterprise platform where businesses can purchase Claude-powered tools from third-party partners. Enterprises with existing Anthropic spend commitments can redirect part of their budget toward partner applications, receiving a single consolidated invoice. Launch partners included Snowflake, GitLab, Harvey, Replit, Rogo, and Lovable. Anthropic does not take a revenue cut from Marketplace purchases, distinguishing it from cloud marketplace models operated by AWS and Microsoft Azure. The Marketplace launched in limited preview, with plans to expand the partner catalog over time.[74]
Launched on April 17, 2026 by Anthropic Labs as a research preview for Pro, Max, Team, and Enterprise subscribers, Claude Design lets users collaborate with Claude (powered by Opus 4.7) to create prototypes, slides, one-pagers, and other visual work through conversation. Outputs can be refined inline or via custom sliders and automatically conform to a team's design system. Exports include Canva, PDF, PPTX, and standalone HTML. Designs can also be shared as internal URLs within an organization or saved as folders. Figma's publicly traded shares fell approximately 7% the day of the announcement. Canva integrated its Design Engine and Visual Suite into Claude Design later in April under a partnership announced at the Sydney office opening.[87][91]
Announced on April 28, 2026, Claude for Creative Work is a coordinated set of nine MCP connectors for professional creative software, available across all Claude plans. The Adobe connector exposes more than 50 tools across Photoshop, Lightroom, Illustrator, Firefly, Premiere, Express, InDesign, and Adobe Stock. Other connectors cover Autodesk Fusion (text-to-3D-modeling), Blender (natural-language Python API access), Ableton, and Splice. Anthropic also announced educational partnerships with the Rhode Island School of Design, Ringling College of Art and Design, and Goldsmiths, University of London.[92]
Unveiled on May 5, 2026 at an invite-only briefing in New York, Claude for Financial Services bundles ten pre-built agent templates aimed at the most time-consuming work in financial institutions: pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, valuation reviewer, general ledger reconciler, month-end closer, statement auditor, and KYC screener. New connectors and MCP apps were announced from Dun & Bradstreet, Fiscal AI, Financial Modeling Prep, Guidepoint, IBISWorld, SS&C Intralinks, Third Bridge, Verisk, and Moody's. Microsoft 365 add-ins for Excel, PowerPoint, Word, and (in beta) Outlook were extended to support Claude. The keynote featured Dario Amodei and JPMorgan Chase Chairman and CEO Jamie Dimon, who demonstrated Claude Code building a Treasury bid-ask spread dashboard in 20 minutes.[95]
Launched on May 12, 2026, Claude for Legal includes 12 practice-area plugins (Commercial Legal, Corporate Legal, Employment Legal, Privacy Legal, Product Legal, Regulatory Legal, AI Governance Legal, IP Legal, and Litigation Legal among others) and more than 20 MCP connectors for legal-specific systems such as DocuSign, Ironclad, and Datasite. Anthropic disclosed that legal professionals had become the most engaged Claude Cowork users of any knowledge-work function.[98]
Launched on May 13, 2026, Claude for Small Business runs on Claude Cowork and connects Claude to QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365 with built-in workflows for finance, operations, sales, marketing, HR, and customer service. The release included 15 ready-made skills covering tasks like payroll planning, bookkeeping, and employee onboarding. It is paired with a multi-city free AI Fluency Tour (Chicago, Tulsa, Dallas, Hamilton Township, Baton Rouge, Birmingham, Salt Lake City, Baltimore, San Jose, Indianapolis) and a free "AI Fluency for Small Business" online course produced with PayPal.[100][101]
A beta feature first released in October 2024 alongside Claude 3.5 Sonnet and Haiku that enables Claude to take screenshots, click, and type text, allowing it to interact with computer interfaces.[52] In February 2026, Anthropic acquired Vercept, a Seattle-based startup specializing in computer-use AI agents, to further advance these capabilities. The Vercept team brought expertise in building agents that can operate full desktop environments, including navigating spreadsheets and managing workflows across multiple tools.[66]
On March 23, 2026, Anthropic launched Computer Use for Mac as a research preview, enabling Claude to control a Mac computer directly. Claude can move the mouse, use the keyboard, open applications, navigate browsers, fill in spreadsheets, and complete multi-step tasks autonomously. When a supported connector is available (such as Google Workspace or Slack), Claude prioritizes that integration but falls back to screen-based control when no connector exists. The feature uses a permission-first approach: Claude requests access before interacting with new applications, and users can halt operations at any point. Computer Use for Mac is available to Pro and Max subscribers and integrates with Dispatch for remote task assignment from a phone. Windows and Linux support has not yet been announced.[78]
In November 2024, Anthropic named AWS as its primary training partner and primary cloud provider. Key aspects of the partnership include:
Anthropic uses AWS Trainium and Inferentia chips for training and deploying models
AWS customers access Claude models through Amazon Bedrock
Amazon's prior $8 billion investment was joined in April 2026 by a new $5 billion commitment at a $350 billion valuation and a separate AWS compute agreement granting Anthropic up to 5 gigawatts of capacity for training and deploying Claude, with Amazon indicating plans to inject up to $20 billion more over time.[19][89]
Anthropic's models are available through Google Cloud's Vertex AI platform, providing enterprise customers with access to Claude capabilities within Google's cloud infrastructure. Google's initial $2 billion investment was joined in April 2026 by a commitment of up to $40 billion ($10 billion initial at a $350 billion valuation, with $30 billion contingent on performance milestones).[14][90]
In September 2025, Claude models were integrated into Microsoft 365 Copilot. In early 2026, the partnership deepened: Anthropic committed to spending $30 billion on Microsoft's Azure cloud services, while Microsoft agreed to invest up to $5 billion in Anthropic. Claude Sonnet 4.6 powers Microsoft Copilot Cowork, a product offering AI agents within the Microsoft 365 suite. Microsoft spends approximately $500 million per year on Anthropic's models. In May 2026, Claude shipped as native Microsoft 365 add-ins for Excel, PowerPoint, and Word at general availability, with an Outlook add-in in public beta. The four add-ins share a single Claude conversation thread per user, allowing context to flow across applications.[47][97]
Announced on May 6, 2026, the SpaceX deal gives Anthropic access to more than 300 megawatts of new compute capacity (over 220,000 NVIDIA GPUs) at SpaceX's Colossus One data center. The capacity expansion allowed Anthropic to double the five-hour Claude Code rate limit for Pro, Max, Team, and seat-based Enterprise subscribers and to raise API rate limits sharply for Tier 1 customers.[96]
In November 2024, Anthropic partnered with Palantir Technologies and Amazon Web Services to provide Claude models to U.S. intelligence and defense agencies for use in classified environments.[53]
Announced on February 17, 2026, the collaboration with Infosys integrates Claude models and Claude Code with Infosys Topaz AI offerings to help enterprises automate complex workflows and accelerate software delivery in regulated industries. The partnership focuses on agentic AI, using the Claude Agent SDK to build persistent agents that handle multi-step tasks such as processing claims, generating and testing code, and managing compliance reviews. The initial focus is on telecommunications, with plans to expand into financial services, manufacturing, and software development. A dedicated Anthropic Center of Excellence supports the collaboration.[71]
Announced on April 24, 2026, the NEC partnership designated NEC Corporation as Anthropic's first Japan-based global partner. Claude will be deployed to approximately 30,000 NEC Group employees worldwide. Joint workstreams cover co-developing AI products for the Japanese market (starting with finance, manufacturing, and local government), integrating Claude into NEC's Security Operations Center, embedding Opus 4.7 and Claude Code into NEC's BluStellar Scenario consulting offerings, and establishing an NEC Center of Excellence with technical enablement and training from Anthropic.[88]
Anthropic and PwC announced a major expansion of their alliance on May 14, 2026, structured around agentic technology build, AI-native deal-making, and reinvention of the enterprise function. PwC will train and certify 30,000 professionals on Claude, roll out Claude Code and Cowork starting with U.S. teams and expanding globally, and stand up a new Finance Business Group using Claude, Claude Cowork, and Claude Code. Across production deployments, PwC reported delivery improvements of up to 70%.[103]
On May 4, 2026, Anthropic, Blackstone, Hellman & Friedman, and Goldman Sachs unveiled a new AI-native enterprise services company to embed Claude into mid-sized companies (initially community banks, manufacturers, regional health systems, and PE portfolio companies). The standalone firm is also backed by General Atlantic, Leonard Green, Apollo Global Management, GIC, and Sequoia Capital, and joins the Claude Partner Network. CNBC reported the venture as a $1.5 billion firm. Anthropic engineering and partnership resources are embedded directly in the new entity.[93][94]
Announced on May 14, 2026, the Anthropic-Gates Foundation partnership commits $200 million over four years in grant funding, Claude usage credits, and technical support across global health, life sciences, education, and economic mobility. The largest pillar focuses on accelerating vaccine and therapy development in low- and middle-income countries. Additional pillars include AI-powered tutoring for U.S. K-12 students; foundational literacy and numeracy programs in sub-Saharan Africa and India; and agricultural productivity tools for the nearly two billion people who depend on smallholder farming. The partners agreed to release agriculture-specific Claude improvements, African-language datasets, and benchmarks as public goods.[102]
Launched in March 2026 with a $100 million investment, the Claude Partner Network supports consulting firms, system integrators, and technology companies that build on Claude for their enterprise customers. Anchor partners include Accenture, Deloitte, Cognizant, Infosys, and PwC; the May 2026 Blackstone/Hellman & Friedman/Goldman Sachs joint venture also joined the network. The program provides training, dedicated technical support, joint go-to-market investment, and a new Claude Certified Architect certification. Anthropic is scaling its partner-facing team fivefold to support the network.[69][93]
Anthropic has experienced rapid enterprise adoption, growing from under 1,000 business customers two years ago to over 300,000 as of 2025.[54] The company has expanded its global presence with offices in San Francisco (headquarters), Dublin, London, Tokyo, Bengaluru, Seoul, Zurich, Sydney (officially opened April 2026), and announced openings in Paris and Munich. EMEA has been Anthropic's fastest-growing region, with run-rate revenue having grown more than 9x year-over-year and the number of large EMEA accounts (over $100,000 ARR) having grown more than 10x. In April 2026, Anthropic announced a London expansion to office space for up to 800 employees in the Knowledge Quarter area, and the company has indicated continued hiring in Dublin from a base of roughly 60 employees in Ireland.[67][91][105]
Major enterprise customers and partners include:
Pfizer: Uses Claude to accelerate research and reduce operational costs
United Airlines: Used Claude to personalize customer messages and improve response speeds
Zoom: Integrated Claude for various business applications
Snowflake: Utilizes Claude for data analytics
Thomson Reuters: CoCounsel tax platform uses Claude for tax professionals
Novo Nordisk: Reduced clinical study report writing from 12 weeks to 10 minutes
Commonwealth Bank of Australia: Reduced customer scam losses by 50%
Rakuten: Cut feature development time by 79% using Claude Code; early Claude Managed Agents customer
Replit: Integrated Claude into Agent for code generation
Canva: Uses Claude for AI-powered design features; integrated Canva Design Engine and Visual Suite into Claude Design in April 2026
NEC: Deploying Claude to ~30,000 employees worldwide
JPMorgan Chase: CEO Jamie Dimon publicly demoed Claude Code at Anthropic's May 2026 financial services briefing[55][88][91][95]
| Year | Annualized Revenue (Run-Rate) |
|---|---|
| 2022 | $10 million |
| 2023 | $100 million |
| 2024 | $1 billion |
| End of 2025 | ~$9 billion |
| February 2026 | $14 billion |
| March 2026 | $19+ billion |
| May 2026 | ~$47 billion |
The company added $6 billion in run-rate revenue during February 2026 alone, driven largely by Claude Code adoption and enterprise growth. By May 2026, alongside the Series H round, Anthropic disclosed that run-rate revenue had crossed $47 billion. The number of customers spending over $100,000 annually has grown 7x year-over-year, and over 500 customers now spend more than $1 million annually, up from roughly a dozen two years ago. Anthropic was capturing over 73% of all spending among companies buying AI tools for the first time as of March 2026. The company has set an internal target of $20 billion to $26 billion ARR for 2026, with a longer-term projection of up to $70 billion in revenue and $17 billion in cash flow by 2028.[1]
Anthropic published a Responsible Scaling Policy (RSP) in September 2023 that establishes safety levels for AI systems based on their capabilities. Models are classified into different Anthropic Safety Levels (ASL), with escalating controls as model risk increases. The policy has been updated multiple times: Version 2.0 in October 2024 introduced new capability thresholds and safety case methodologies; Version 2.1 in March 2025 added CBRN-related thresholds; Version 2.2 in May 2025 coincided with ASL-3 activation for Claude 4 models; and Version 3.0 on February 24, 2026 introduced Frontier Safety Roadmaps with detailed safety goals, Risk Reports that quantify risk across all deployed models, and a restructured two-tier mitigation framework distinguishing unilateral commitments from goals contingent on industry cooperation. Anthropic acknowledged that scaling protections to ASL-4 and beyond cannot be done by any single firm in isolation, particularly against state-level adversaries. ASL-3 remained the operative posture for Anthropic's frontier models as of May 2026; ASL-4 capability thresholds are defined in the RSP but had not been activated.[56]
Key research focus areas include:
AI Alignment: Ensuring AI systems behave as intended
Mechanistic Interpretability: Understanding how AI systems make decisions
Constitutional AI: Developing value-aligned AI systems
Safety Research: Mitigating risks from advanced AI systems
Deceptive Behavior Studies: Research on "sleeper agent" behaviors that can persist through safety fine-tuning[57]
Launched on March 11, 2026, the Anthropic Institute is a dedicated research arm focused on studying AI's societal impact. Led by co-founder Jack Clark (in a new role as Head of Public Benefit), the Institute brings together three teams: the Frontier Red Team (which stress-tests AI systems at their capability limits), Societal Impacts (which studies real-world AI usage), and Economic Research (which tracks AI's effects on employment and the broader economy). The Institute also incubates new research efforts, including forecasting AI progress and studying how advanced AI will interact with the legal system. Notable recruits include Matt Botvinick (formerly of Google DeepMind), Anton Korinek (University of Virginia), and Zoe Hitzig (previously at OpenAI).[68]
Launched on April 7, 2026, Project Glasswing is a cybersecurity coalition founded by Anthropic alongside AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. The coalition uses access to Anthropic's restricted Claude Mythos Preview model to identify and disclose vulnerabilities in critical software before similar capabilities become widely accessible. Anthropic committed up to $100 million in Mythos Preview usage credits to Glasswing participants and approved organizations, $2.5 million to Alpha-Omega and OpenSSF through the Linux Foundation, and $1.5 million to the Apache Software Foundation. Mythos Preview is not publicly released.[82][83]
Red Teaming: Regular testing for vulnerabilities and potential misuse
External Audits: Collaboration with organizations like the US AI Safety Institute and UK Safety Institute
Constitutional Classifiers: Defenses against jailbreaking attempts, with the second-generation system (Constitutional Classifiers++) reducing compute overhead to ~1% while maintaining strong jailbreak resistance[40]
Cyber Verification Program: Introduced with Opus 4.7 in April 2026, granting vetted security researchers legitimate access for authorized research alongside automated detection of prohibited cyber requests in the production API.[86]
In October 2023, Anthropic was sued by music publishers including Universal Music Group, Concord, and ABKCO for alleged copyright infringement of song lyrics used in training data. A judge later denied a request for a preliminary injunction while litigation continued.[58] On January 29, 2026, the publishers amended their complaint to seek up to $3 billion in damages over alleged BitTorrent downloads of 20,517 compositions including lyrics and sheet music. As of May 2026 the case has not settled; Anthropic moved for summary judgment on fair-use grounds in April 2026 and motion practice is ongoing in the Middle District of Tennessee.[106]
In September 2025, Anthropic agreed to pay $1.5 billion to settle the Bartz v. Anthropic class action lawsuit brought by authors who alleged the company had illegally downloaded millions of pirated books from shadow libraries like Library Genesis and Pirate Library Mirror to train its AI models. The settlement, the largest copyright settlement in U.S. history, covers approximately 500,000 works at roughly $3,000 per book. Judge William Alsup had earlier ruled on summary judgment that using books to train AI was fair use if acquired legally, but denied Anthropic's motion regarding pirated copies.[27] Class counsel filed a Motion for Final Approval of Settlement on March 19, 2026, ahead of a March 30, 2026 claims deadline. By that deadline, 440,490 of 482,460 eligible works had been claimed, a 91.3% claim rate. The final approval hearing was scheduled for May 14, 2026; payouts are expected after final approval and resolution of any appeals.[107]
On February 24, 2026, Anthropic published a detailed blog post accusing three Chinese AI companies of conducting industrial-scale distillation attacks against Claude. The companies, DeepSeek, Moonshot AI, and MiniMax, allegedly created over 24,000 fraudulent accounts and generated more than 16 million exchanges with Claude to train their own models. The attackers used commercial proxy services and what Anthropic called "hydra cluster" networks, with one proxy setup controlling more than 20,000 fraudulent accounts at once. Anthropic argued that distillation-derived models strip away safety protections, creating national security risks. The accusations drew mixed reactions: some observers supported Anthropic's concerns about intellectual property theft, while others noted the irony given Anthropic's own legal challenges over training data sourced from copyrighted works.[65]
In early 2026, Anthropic's relationship with the U.S. Department of Defense deteriorated after the Trump administration demanded that AI companies remove all company-specific guardrails from military contracts. Anthropic refused to allow unrestricted use of its models, insisting on retaining two prohibitions: against mass domestic surveillance and against fully autonomous weapons systems. On March 3, 2026, the Department of Defense formally designated Anthropic a supply chain risk to national security, the first time such a designation had been applied to an American company. On March 9, Anthropic filed lawsuits in two federal courts (the U.S. District Court for the Northern District of California and the U.S. Court of Appeals for the Federal Circuit) challenging the designation as an unconstitutional retaliation for protected speech. Nearly 150 retired federal and state judges subsequently filed an amicus brief supporting Anthropic. Microsoft, Google engineers, and employees at competing AI companies also voiced support. The financial impact could reach billions of dollars in lost government revenue. OpenAI subsequently signed a contract with the Defense Department to replace Anthropic's services.[33]
The company's partnerships with Amazon and Google have attracted regulatory attention:
The UK Competition and Markets Authority investigated Amazon's partnership but concluded it could not be examined under current merger rules
The Federal Trade Commission has reviewed Big Tech AI investments but has not taken enforcement action
The dramatically enlarged Amazon and Google commitments announced in April 2026 (up to $25 billion and $40 billion respectively in cash and compute, alongside the corresponding cloud revenue flowing back from Anthropic) have intensified attention to the "circular" structure of AI-related earnings, with Fortune reporting that mark-to-market gains on the Anthropic stake accounted for a large share of headline AI profits at both companies during the period.[59][89][90]
The leadership team is composed of experts with experience across leading tech companies and research institutions:
| Name | Title | Background |
|---|---|---|
| Dario Amodei | Co-founder and CEO | Former VP of Research at OpenAI; PhD in Computational Neuroscience from Princeton University[60] |
| Daniela Amodei | Co-founder and President | Previously held roles at OpenAI and Stripe |
| Rahul Patil | Chief Technology Officer | Former CTO of Stripe; joined October 2025[63] |
| Sam McCandlish | Co-founder and Chief Architect | Previously worked on GPT-3 at OpenAI; transitioned from CTO to Chief Architect in October 2025[63] |
| Mike Krieger | Co-lead of Labs | Co-founder of Instagram; joined as CPO in May 2024, moved to Labs co-lead in late 2025[61] |
| Ami Vora | Head of Product | Succeeded Mike Krieger as Head of Product in late 2025[63] |
| Jack Clark | Co-founder, Head of Public Benefit | Former Policy Director at OpenAI; leads the Anthropic Institute |
| Tom Brown | Co-founder, Head of Core Resources | Previously worked on GPT-3 at OpenAI and at Google DeepMind |
| Krishna Rao | Chief Financial Officer | Former head of corporate finance at Airbnb |
| Paul Smith | Chief Commercial Officer | First CCO appointment |
| Jason Clinton | Chief Information Security Officer | Former Staff Software Engineer at Google |
| Jan Leike | Co-lead of Alignment Science | Former alignment researcher at OpenAI |
| Chris Ciauri | Managing Director of International | Former President of EMEA for Google Cloud; 25 years in tech including a decade at Salesforce |
| Sarah Heck | Head of Public Policy | Joined as Head of External Affairs, promoted to lead expanded Public Policy organization[68] |
| Irina Ghose | Managing Director, India | Leads the Bengaluru office; India is Claude's second-largest market[71] |
| Theo Hourmouzis | General Manager, Australia & New Zealand | Joined from Snowflake (SVP, ANZ & ASEAN); leads Sydney office[91] |
Anthropic competes in a rapidly evolving AI market with both Western and Chinese rivals. As of early-to-mid 2026, the competitive landscape includes:
OpenAI: Creator of ChatGPT and GPT models, including GPT-5.4 released in March 2026. Anthropic's most direct competitor in both consumer and enterprise AI, and its successor for the Pentagon contract after Anthropic's departure. In May 2026, Anthropic's $965 billion Series H valuation overtook OpenAI's $852 billion (set in a $122 billion March 2026 round) to make Anthropic the most valuable AI startup.
Google DeepMind: Developer of Gemini models. Also an investor in Anthropic through Google's increasingly large stake.
Meta: Developer of open-source Llama models, which compete on cost and accessibility.
xAI: Elon Musk's AI company, developer of the Grok model family. Competes aggressively on pricing, with Grok 4.1 offering a 2 million token context window at very low cost.
DeepSeek: Chinese AI lab that has driven prices down significantly while achieving competitive benchmark scores. Subject of Anthropic's distillation accusations in February 2026.
Mistral AI: European AI startup focused on efficient models.
Anthropic has differentiated itself through its safety-first approach, strong coding performance (particularly through Claude Code and Claude Opus 4.7), and enterprise adoption. By March 2026, the company was capturing over 73% of first-time AI tool spending among enterprises, up from a 50/50 split with OpenAI just 10 weeks earlier.[1]
AI safety
Amazon Bedrock
Vertex AI
Several Anthropic-specific topics now have dedicated pages: