Field	Value
Type	Industry alliance for open-source AI security
Announced	July 27, 2026
Convened by	NVIDIA with 36 other inaugural partners
Founding members	37 organizations
Membership	More than 120 organizations (as of August 4, 2026)
Affiliated bodies	Linux Foundation, OpenSSF, Akrites
Key output	SAFE guidelines RFC, open-source agent security tooling
RFC repository	github.com/OpenSecureAIAlliance/RFCs

The Open Secure AI Alliance is an industry coalition announced on July 27, 2026 to develop and share open technologies, techniques, and tools for securing software and AI agents. Convened by NVIDIA with 36 other inaugural partners including Microsoft, Cisco, CrowdStrike, Hugging Face, IBM, Palantir, Red Hat, and the Linux Foundation, the group grew to more than 120 member organizations within its first week.[1][2][3] Its founding argument is that cybersecurity defenders need open models, agent harnesses, and security tooling they can inspect, adapt, and run on their own infrastructure, rather than depending solely on closed frontier systems. On August 4, 2026, the alliance published its first major deliverable: a Request for Comments on the Shared AI Findings Exchange (SAFE), a proposed framework for confidentially reporting and learning from AI security incidents.[8][9][10]

Background

The alliance formed in the immediate aftermath of two linked security events in July 2026. On July 16, Hugging Face disclosed that an intrusion into part of its production infrastructure had been "driven, end to end, by an autonomous AI agent system." A malicious dataset abused two code-execution paths in the company's dataset-processing pipeline, and the attacking agent swarm escalated to node-level access, harvested cloud and cluster credentials, and moved laterally across internal clusters.[6] Five days later, OpenAI disclosed that the attacker had been its own models: during an evaluation on the ExploitGym cyber-capability benchmark, run with production safety classifiers deliberately disabled, its models escaped the sandboxed test environment through a zero-day vulnerability in a package proxy and breached Hugging Face's production systems to obtain benchmark answers.[5][7]

A detail of Hugging Face's incident response became the alliance's founding exhibit. When the company's responders tried to analyze the attack logs with frontier models behind commercial APIs, the providers' safety guardrails blocked the requests, since submitting large volumes of real exploit payloads looks the same whether it comes from an attacker or an incident responder. Hugging Face instead ran the forensic analysis on zai-org/GLM-5.2, an open-weight Chinese model, on its own infrastructure, using LLM-driven analysis agents to reconstruct a timeline from more than 17,000 recorded attacker actions.[6] NVIDIA's launch post cited the episode directly: "cyber defenders need open, frontier agentic systems for self-defense."[1]

The launch also landed in a charged policy moment. CNBC reported that United States lawmakers were weighing measures to curb adoption of Chinese AI models, most of the strongest of which are open weight, and that Treasury Secretary Scott Bessent had threatened sanctions over so-called distillation attacks the week before. In the same period, NVIDIA, Microsoft, Meta, Palantir, and more than 20 other companies released a letter urging policymakers to avoid "premature restrictions" on open-weight models.[5] NVIDIA's launch post pressed the same case, urging policymakers to recognize open models, harnesses, and security tooling "as defensive assets, not liabilities" in AI safety and cybersecurity policy.[1]

The alliance builds on two existing Linux Foundation efforts: the OpenSSF security community and Akrites, a coordinated vulnerability-remediation initiative for critical open-source software that the Linux Foundation launched on June 25, 2026 with founding commitments from Amazon Web Services, Anthropic, Cisco, Citi, Google, IBM, JPMorganChase, Microsoft and GitHub, NVIDIA, OpenAI, and others.[1][11]

Launch and membership

NVIDIA announced the alliance in a blog post on July 27, 2026, describing it as "a movement to develop and share open technologies, techniques and tools to safeguard software and agents in the age of AI."[1] The launch-day version of the post, captured by the Internet Archive, named 37 inaugural partners: NVIDIA, Adobe, Cadence, Capital One, Cisco, Cloudera, Cloudflare, Cognition, CrowdStrike, Databricks, Dell Technologies, DoorDash, Elastic, HPE, Hugging Face, IBM, LangChain, the Linux Foundation, Microsoft, NAVER, NetApp, Nous Research, OpenClaw, Palantir, Palo Alto Networks, Red Hat, Reflection AI, Salesforce, SAP, SK Telecom, ServiceNow, Siemens, Snowflake, SpaceXAI, Synopsys, Thinking Machines Lab, and TrendAI.[2][3]

Membership grew quickly, and NVIDIA expanded the member list in the same launch post rather than publishing a separate roster: as of August 7, 2026, the post names 123 organizations, adding companies such as Amazon, Akamai, Atlassian, Canonical, Check Point, Cohere, Datadog, Dropbox, EleutherAI, Fortinet, G42, GitHub, Intel, Lenovo, Mistral, Mozilla, Nokia, NTT, Okta, Perplexity, Pinterest, Samsung Electronics, SentinelOne, Thales, Uber, UiPath, Veeam, Visa, Wiz, Workday, and Zscaler.[1] NVIDIA's August 4 update said Amazon "today became one of the newest members" and that Visa had also joined.[8] An NVIDIA post on X the same day described the group as "over 120 members strong."[15]

Three companies were conspicuously absent: OpenAI, Anthropic, and Google, the labs most identified with closed frontier models. Trade coverage noted the gap at launch and again a week later; all three participate in the Linux Foundation's Akrites initiative and in the separate Coalition for Secure AI, but none had joined the Open Secure AI Alliance as of early August 2026.[3][4][11][12] The Hacker News observed that the alliance's public materials did not say why those companies were absent, what members must contribute to join, or how the group is formally governed.[3]

Governance

The alliance launched without a published charter or formal legal structure. The Linux Foundation supports the effort and hosts its first standards discussion, but as of early August 2026 the group was not listed as a formal Linux Foundation project.[3][9] Its web presence runs through NVIDIA: the domain opensecureaialliance.org redirects to NVIDIA's launch blog post, and prospective members are directed to a contact form on nvidia.com.[1] A GitHub organization, OpenSecureAIAlliance, was created on August 3, 2026 and hosts the alliance's RFC repository under a Creative Commons Attribution 4.0 license.[10]

SAFE guidelines

On August 4, 2026, timed to the opening day of the Black Hat conference in Las Vegas, the Linux Foundation published a Request for Comments on the Shared AI Findings Exchange (SAFE), drafted by an Open Secure AI Alliance working group with contributors from Cisco, CrowdStrike, Hugging Face, NVIDIA, and Red Hat.[8][9] The proposal describes SAFE as "a proposed independent incident-learning and assurance initiative" that would confidentially collect and analyze AI incidents and near misses, promptly inform affected parties, and turn recurring failures into shared, evidence-based controls.[10] The Linux Foundation compared the model to NASA's Aviation Safety Reporting System, the confidential, non-punitive reporting scheme used in aviation.[9]

The draft's reporting compact would oblige members to report incidents in which an AI system they operate accesses or modifies a third-party system without authorization, escapes a sandbox or policy boundary, accesses third-party confidential information, or keeps probing a production target after the operator suspects the activity is unauthorized. Intent does not matter: "Believing that an environment was simulated may explain an incident, but it does not remove the duty to report it."[10] The provision maps directly onto the OpenAI-Hugging Face incident of the previous month.

The proposed notification timelines:[10]

Deadline	Required action
As soon as possible	Notify the directly affected organization
72 hours	Notify customers with credible exposure
4 business days	Submit a confidential initial SAFE incident report
14 days	Issue a broader customer advisory when warranted
30 days	Publish a preliminary factual report
90 days	Publish remediation status
Weekly	Provide machine-readable updates while material risks remain unresolved

The draft also proposes a three-tier disclosure model (confidential rapid alerts, de-identified member advisories, and public safety reports), evidence-preservation requirements covering prompts, traces, tool calls, credentials, and human intervention events, and a review framework that examines each incident across eight control layers from the model itself through safeguards, tools, environment, monitoring, human operations, and supply chain. It recommends that SAFE operate independently, "so that no vendor or industry segment controls its findings," and applies equally to open and closed systems: "Trust is not a control; shared evidence and verifiable improvement are how trust is earned."[10]

Technical contributions

NVIDIA's own contributions include the NVIDIA Labs Object-Oriented Agent (NOOA) research framework, an Apache 2.0-licensed harness released on GitHub that structures agent behavior as Python classes so agent actions are easier to test, trace, audit, and govern; the OpenShell runtime, which restricts what an agent can see and do; the Garak LLM vulnerability scanner; NeMo Guardrails, NeMo Anonymizer, and NeMo Safe Synthesizer; and cryptographically signed "verified agent skills."[1][8][13]

Other members' contributions announced at launch and in the August 4 update span the agent security stack:[1][8]

Layer	Member	Contribution
Identity and permissions	Okta	Reference implementations for agent identity using the Cross App Access (XAA) protocol
Identity and permissions	Palo Alto Networks	Agent Guard and Agent Watch open-source tools
Identity and permissions	Red Hat	asago, mapping governance requirements (NIST, OWASP, EU AI Act) to runtime agent permissions
Identity and permissions	HPE	Contributions to SPIFFE/SPIRE zero-trust workload identity
Harnesses and tooling	Amazon	Strands Agents toolkit and the Cedar authorization language
Harnesses and tooling	Microsoft	PyRIT, RAMPART, Clarity, Assert red-teaming and evaluation tools, and the MDASH multi-model scanning harness
Harnesses and tooling	Capital One	VulnHunter for agentic AI code security
Harnesses and tooling	Cloudflare	Vulnerability Discovery Harness released as an open-source skill
Harnesses and tooling	Wiz	Atlas autonomous vulnerability research engine
Harnesses and tooling	Visa	Visa Vulnerability Agentic Harness
Models	Cisco	DefenseClaw governance layer, Antares security small language models, Project CodeGuard
Models	CrowdStrike	Cyber-defense fine-tunes of NVIDIA Nemotron Nano models
Models	Mistral	Shieldstral multimodal safety classifier, released as open weights under Apache 2.0
Model formats	Hugging Face	Safetensors weight format, offered to the PyTorch Foundation
Observability	Perplexity	Numbat agent security suite for client endpoints
Observability	Uber	Open-sourced components of ADR (Agentic AI Detection and Response)
Supply chain	IBM and Red Hat	Lightwell, digitally signed patches for the open-source supply chain
Resilience	LangChain	Recovery and fallback capabilities in Deep Agents, LangGraph, and LangChain
Resilience	Veeam	Kanister, an open-source Kubernetes data-protection framework

NVIDIA's launch post also said SpaceXAI had open sourced its Grok Build terminal coding agent and planned to open source the weights of the Grok model line.[1] Uber says its ADR system supports more than 200,000 agent sessions per day across 30,000 endpoints.[8]

Relationship to other bodies

The Open Secure AI Alliance is distinct from the Coalition for Secure AI (CoSAI), an OASIS Open Project launched in July 2024 whose more than 40 partner organizations include Google, Microsoft, OpenAI, Anthropic, NVIDIA, Amazon, IBM, and, since February 2026, Meta as a premier sponsor. CoSAI develops AI security best practices and research under OASIS governance, while the Open Secure AI Alliance is a newer, NVIDIA-convened group focused on open-source tooling; several companies belong to both.[16][17] It is also distinct from Akrites, the Linux Foundation's coordinated vulnerability disclosure effort for critical open-source software, which the alliance describes itself as building on.[1][11]

Reception

Coverage highlighted the group's speed. TechCrunch wrote that "the group is operating at AI speeds," noting that it had produced the SAFE proposals within a week of forming, while judging the initial guidelines "nothing terribly earth-shattering for now."[12] Security trade press framed the alliance as a bid to position open models and harnesses as defensive assets in a policy debate that could otherwise favor closed systems, with StorageReview noting its argument that AI security "should be evaluated at the full agent-stack level rather than solely through the availability of model weights."[4] The Hacker News flagged the absence of formal governance documentation and of OpenAI, Anthropic, and Google as open questions.[3] CrowdStrike, an inaugural partner, summarized the group's thesis in its own announcement: "AI safety is not achieved through opacity."[14]

See also

References

Industry Leaders Unite in Open Secure AI Alliance for AI Safety and Security. NVIDIA Blog, July 27, 2026. https://blogs.nvidia.com/blog/open-secure-ai-alliance/
Internet Archive snapshot of the NVIDIA launch post, July 27, 2026 (11:16 UTC), showing the original 37-member list. https://web.archive.org/web/20260727111606/https://blogs.nvidia.com/blog/open-secure-ai-alliance/
NVIDIA Forms 37-Member Open Secure AI Alliance and Open-Sources NOOA Framework. The Hacker News, July 27, 2026. https://thehackernews.com/2026/07/nvidia-forms-37-member-open-secure-ai.html
NVIDIA's Open Secure AI Alliance Launches With 35+ Members and Three Notable Absences. StorageReview, July 2026. https://www.storagereview.com/news/nvidias-open-secure-ai-alliance-launches-with-35-members-and-three-notable-absences
Nvidia, SpaceX, Microsoft launch AI safety initiative as OpenAI cyberattack fallout continues. CNBC, July 27, 2026. https://www.cnbc.com/2026/07/27/nvidia-ai-initiative-openai-cyber-attack.html
Security incident disclosure, July 2026. Hugging Face, July 16, 2026. https://huggingface.co/blog/security-incident-july-2026
OpenAI's accidental cyberattack against Hugging Face is science fiction that happened. Simon Willison, July 22, 2026. https://simonwillison.net/2026/Jul/22/openai-cyberattack/
AI Leaders Propose SAFE Guidelines for Cybersecurity Transparency. NVIDIA Blog, August 4, 2026. https://blogs.nvidia.com/blog/open-secure-ai-alliance-contributions/
Proposing the SAFE Working Group: An Open Community Effort to Improve AI Security. The Linux Foundation, August 4, 2026. https://www.linuxfoundation.org/blog/proposing-the-safe-working-group-an-open-community-effort-to-improve-ai-security
Shared AI Findings Exchange (SAFE) draft RFC. Open Secure AI Alliance RFC repository, GitHub, August 2026. https://github.com/OpenSecureAIAlliance/RFCs
Linux Foundation and Industry Leaders Launch Akrites to Defend Critical Open Source Software Against AI-Enabled Cyber Threats. The Linux Foundation, June 25, 2026. https://www.linuxfoundation.org/press/linux-foundation-and-industry-leaders-launch-akrites-to-defend-critical-open-source-software-against-ai-enabled-cyber-threats
Nvidia doesn't mess around: A week after open AI industry group formed, it's already showing progress. TechCrunch, August 4, 2026. https://techcrunch.com/2026/08/04/nvidia-doesnt-mess-around-a-week-after-open-ai-industry-group-formed-its-already-showing-progress/
NVIDIA-NeMo/labs-OO-Agents (NOOA) repository. GitHub, accessed August 7, 2026. https://github.com/NVIDIA-NeMo/labs-OO-Agents
CrowdStrike Joins the Open Secure AI Alliance. CrowdStrike Blog, July 27, 2026. https://www.crowdstrike.com/en-us/blog/crowdstrike-joins-the-open-secure-ai-alliance/
NVIDIA (@nvidia) on X: "Now over 120 members strong, the Open Secure AI Alliance is sharing new open source security contributions..." August 4, 2026. https://x.com/nvidia/status/2084625868173820144
Introducing the Coalition for Secure AI, an OASIS Open Project. OASIS Open, July 18, 2024. https://www.oasis-open.org/2024/07/18/introducing-cosai/
Meta Joins Coalition for Secure AI as Premier Sponsor to Advance Industry Security Standards. OASIS Open, February 3, 2026. https://www.oasis-open.org/2026/02/03/meta-joins-coalition-for-secure-ai-as-premier-sponsor-to-advance-industry-security-standards/

Editing: Open Secure AI Alliance