AI in cryptocurrency
Last reviewed
Sources
No citations yet
Review status
Needs citations
Revision
v3 ยท 3,948 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
No citations yet
Review status
Needs citations
Revision
v3 ยท 3,948 words
Add missing citations, update stale details, or suggest a clearer explanation.
AI in cryptocurrency refers to the use of artificial intelligence techniques, including machine learning, large language models, and AI agents, within cryptocurrency and blockchain systems. The two fields intersect in several distinct ways: AI is used to trade, analyze, and secure crypto markets; blockchains are proposed as infrastructure for decentralized AI compute, data, and provenance; a category of tokens and projects markets itself around an "AI" narrative; and generative AI has become a tool for crypto-related fraud. The relationship is heavily promoted and frequently overhyped, and many "AI" crypto products are marketing labels rather than functioning autonomous systems [1]. This article describes the documented applications, projects, and risks, and attributes promotional claims to their sources rather than endorsing them.
This page focuses on the crypto and blockchain specific intersection. For AI in banking, lending, insurance, and traditional capital markets, see AI in finance, which covers algorithmic trading on regulated exchanges, fraud detection, credit scoring, and compliance in more depth.
Artificial intelligence and cryptocurrency are independent technologies that have converged for both technical and commercial reasons. On the technical side, cryptocurrency markets generate large volumes of public, machine-readable data: every transaction on a public blockchain is visible, exchanges publish high-frequency price and order-book feeds, and social media activity around tokens is abundant. This data is well suited to machine learning, which has driven the use of AI for trading, on-chain analysis, and fraud detection.
A second, more speculative strand treats blockchains as coordination infrastructure for AI itself. Proponents argue that token incentives can be used to assemble decentralized networks of GPU compute, training data, and model inference outside the control of large technology companies, and that cryptographic records can help establish the provenance of AI-generated content. These ideas remain early and unproven at scale, and their economic viability is contested.
A third strand is commercial and narrative-driven. "AI" became a dominant marketing theme in crypto during 2023 to 2025, producing a category of "AI tokens" and "AI agent" tokens whose market values often moved with broader AI sentiment rather than with any deployed product, and commentators have cautioned that much of this activity is speculative and that many projects branded as AI agents are, in practice, chatbots or memecoins with limited autonomy [1].
Finally, the same generative AI tools that have legitimate uses have lowered the cost of crypto fraud. Deepfake video, voice cloning, and automated text generation are now common components of investment scams that use cryptocurrency as the payment rail, a risk documented by law enforcement and blockchain-analytics firms [2][3].
The table below summarizes the main areas where AI and cryptocurrency intersect. Specific projects are discussed and sourced in later sections.
| Area | Description | Maturity |
|---|---|---|
| Automated trading bots | Software that executes crypto trades across exchanges using models, signals, or rules | Established; widely used, quality varies |
| Market prediction and sentiment analysis | Models that ingest price, on-chain, and social data to forecast moves or gauge mood | Established as a tool; predictive value disputed |
| On-chain analytics and AML | Machine learning to trace funds, detect laundering, and flag illicit wallets | Established; used by compliance firms and investigators |
| Smart-contract auditing | AI assistance in finding vulnerabilities in contract code before deployment | Emerging; complements, does not replace, human auditors |
| AI agents with wallets | Autonomous software agents that hold crypto and transact on-chain | Early; infrastructure launched 2024 to 2025 |
| Decentralized AI compute and data | Token-incentivized networks for GPU compute, model inference, and training data | Early; live networks exist, economics unproven |
| Content provenance | Blockchain-anchored records to attest the origin of AI-generated media | Early and niche; competes with non-blockchain standards |
| AI-driven crypto scams | Generative AI used to defraud victims via crypto payment rails | Widespread and growing risk |
Automated trading is the most established use of software in crypto markets. Retail platforms and trading-bot services let users run strategies on centralized and decentralized exchanges around the clock, and quantitative crypto funds apply techniques such as reinforcement learning, time-series models, and execution algorithms to manage positions and reduce slippage. Because crypto markets trade continuously and across many venues, they are a natural fit for automated execution and cross-exchange arbitrage.
Claims about the returns of AI-driven crypto trading should be treated with caution. Performance figures in this space are frequently self-reported, are not independently audited, and suffer from survivorship and selection bias because failed strategies are rarely reported. No trading approach, including those marketed as AI-powered, eliminates the substantial risk of loss, and crypto-specific risks such as exchange failures, smart-contract exploits, and extreme volatility apply regardless of the strategy. For a broader treatment of AI in trading on regulated markets, see algorithmic trading and AI in finance.
A common AI application is sentiment analysis: using natural language processing to score posts on social media, news, and on-chain messages as bullish or bearish for particular tokens. Some products combine this with on-chain signals, such as flows to and from exchanges, to generate trading indicators. Sentiment tooling can summarize large volumes of text quickly, but its predictive value for price is disputed. Crypto social media is also a target for manipulation, including coordinated promotion and bot activity, which can poison the inputs that sentiment models rely on.
One of the most concrete uses of AI in crypto is blockchain analytics for compliance and investigations. Firms such as Chainalysis, Elliptic, and TRM Labs build software that clusters addresses, traces fund flows, and assigns risk scores to wallets and transactions, which exchanges and banks use to meet anti-money-laundering (AML) obligations. Because illicit actors increasingly move funds across many chains, machine learning is used to surface laundering patterns that are difficult for human analysts to find manually [4].
Vendors have begun adding generative AI and agentic features to these tools. Elliptic launched an AI "copilot" in April 2025 that the company says summarizes risk triggers and fund flows, surfaces intelligence on involved entities, and drafts material to support regulatory filings, with human review retained [4]. In March 2026, Chainalysis announced plans for AI-powered blockchain-intelligence agents intended to automate parts of fund tracing, suspicious-activity identification, and compliance workflows [5]. According to Chainalysis, illicit cryptocurrency addresses received at least $154 billion in 2025, which the firm notes still represented less than 1% of attributed on-chain transaction volume [6]. The scale and cross-chain complexity of this activity is a primary driver of AI adoption in compliance.
Smart contracts hold large sums and are frequent targets of attack, so there is strong interest in using AI to find vulnerabilities before deployment. Researchers have built tools that fine-tune large language models, sometimes combined with retrieval and static analysis, to detect issues such as reentrancy, integer overflow, and access-control flaws in contract code.
The published results suggest AI is a useful aid but not a replacement for human auditors. A 2025 framework from Georgia Tech researchers called LLMBugScanner, which combines fine-tuned language models with ensemble voting on Ethereum contracts, reported a top-five detection accuracy of about 60%, roughly 19% higher than individual baseline models; the researchers found no single model performed well across all vulnerability types, that access-control and constructor errors remained hard to detect, and that about 10% of model outputs included invented or unsupported vulnerabilities (hallucinations) [7]. Security companies including OpenZeppelin, Forta, Hexagate, and Hypernative also apply machine learning to real-time monitoring of deployed protocols, aiming to detect exploits in progress and trigger automated responses. Despite these tools, on-chain theft remained severe: Chainalysis estimated about $3.4 billion was stolen through hacks in 2025, including a roughly $1.5 billion theft from the exchange Bybit attributed to North Korean actors, which it described as the largest single crypto heist on record [6].
A newer area is giving autonomous software agents the ability to hold cryptocurrency and transact on-chain. In 2025, Coinbase released AgentKit and related "agentic wallet" infrastructure, which the company describes as giving an AI agent its own wallet and a set of on-chain actions such as transfers, swaps, and smart-contract interactions, with guardrails including transaction limits and key isolation so private keys are not exposed to the agent's prompt or model [8][9]. Proponents argue that crypto rails suit machine-to-machine payments because they settle quickly, operate continuously, and do not require an agent to open a traditional bank account.
Related to this is x402, an open payment standard Coinbase introduced in May 2025 that repurposes the dormant HTTP 402 "Payment Required" status code so that a client, including an AI agent, can pay for a web resource or API call with a stablecoin in a single HTTP exchange rather than using accounts or API keys, with later updates adding multi-chain support and broader governance [10]. These systems are early, and autonomous agents that transact with real funds raise unresolved questions about security, liability, and the consequences of agent errors or manipulation.
Several blockchain projects aim to build decentralized infrastructure for AI using token incentives. The general idea is to aggregate computing power, model inference, or data from many independent providers and coordinate and pay for it on-chain, positioned as an alternative to centralized cloud providers.
Examples include networks that market decentralized GPU compute, such as Akash Network, Render, and io.net, which aggregate hardware from data centers and independent operators; Bittensor, which describes itself as an incentivized network for machine-intelligence outputs (discussed below); and data-focused projects such as Ocean Protocol, which provides tooling for tokenized data marketplaces and "compute-to-data" access intended to let models train on datasets without the raw data leaving the owner's control. Vendor and project comparisons frequently claim that decentralized GPU access undercuts mainstream cloud pricing, but such figures originate from the projects and resellers themselves and depend heavily on hardware type, availability, and reliability, so they should be treated as vendor claims rather than independent benchmarks. The durability and real-world demand for these networks remain unproven.
Generative AI has become a significant tool for crypto-related fraud, which is covered in detail under risks and controversies below. Because cryptocurrency payments are fast and difficult to reverse, crypto is a common payout mechanism for AI-enabled investment scams, fake-endorsement schemes, and social-engineering attacks. This is a documented harm rather than a beneficial application, and it is one of the most consequential ways AI and crypto currently intersect [2][3].
From 2023 onward, "AI" became a leading marketing theme in cryptocurrency, giving rise to a loosely defined category of "AI tokens." These include the native tokens of decentralized-AI infrastructure projects (such as compute and data networks) and, separately, a wave of "AI agent" tokens launched during a late-2024 speculative surge. Market trackers group dozens of assets under an "AI" label, but the grouping is informal and inconsistent, and reported category market values vary widely between sources, so specific figures should be read as approximate and source-dependent.
A prominent example of the agent-token wave was ai16z, a Solana-based project whose name parodied the venture firm Andreessen Horowitz (a16z) and which was associated with an open-source agent framework later rebranded as ElizaOS. Another was Virtuals Protocol, a platform on which users can create AI agents and launch associated tokens. Reporting from the period described these tokens rising sharply alongside agents that posted on social media, including AIXBT, which presents itself as analyzing crypto projects, and meme-driven tokens such as GOAT (Goatseus Maximus) connected to an experiment called Truth Terminal [1].
The AI-token narrative should be described neutrally and not endorsed. Several points are well documented. Token prices in this category have been highly volatile and have frequently moved on sentiment and announcements rather than on deployed, revenue-generating products. Industry participants and journalists have repeatedly questioned whether many "AI agents" are genuinely autonomous; in widely cited examples humans remained in the loop, and critics characterized a large share of agent tokens as "chatbots with memecoins attached" [1]. Holding a project's token is also generally not the same as owning or controlling the underlying technology. This article does not recommend buying, selling, or holding any token, and the projects named here are mentioned because they were widely reported, not as endorsements.
The following projects are frequently cited in the AI-crypto space. Descriptions reflect how the projects and reporting characterize them; claims about performance or scale are attributed and are not independently verified here.
| Project | Stated focus | Notes |
|---|---|---|
| Bittensor (TAO) | Incentivized network for machine-intelligence outputs | Caps supply at 21 million tokens with periodic halvings; restructured emissions via "Dynamic TAO" subnets in February 2025 [11] |
| Render | Decentralized GPU rendering and compute marketplace | Expanded toward AI compute; provider and reseller pricing claims are vendor-sourced |
| Akash Network | Decentralized cloud compute marketplace | Markets GPU access as cheaper than mainstream cloud; figures are vendor-sourced |
| io.net | Aggregated decentralized GPU compute for ML | Pools idle hardware from data centers and operators |
| Ocean Protocol | Tokenized data marketplaces for AI | Part of the Artificial Superintelligence Alliance token merger |
| Fetch.ai | Autonomous "agent" infrastructure | Merged token with SingularityNET and Ocean Protocol under the Artificial Superintelligence Alliance |
| SingularityNET | Decentralized marketplace for AI services | Part of the Artificial Superintelligence Alliance |
| Virtuals Protocol | Platform to create AI agents with associated tokens | Central to the 2024 to 2025 agent-token surge [1] |
| ElizaOS (formerly ai16z) | Open-source multi-agent framework and associated token | Rebranded from ai16z in early 2025 [1] |
| Coinbase AgentKit / x402 | Wallet and payment infrastructure for AI agents | AgentKit gives agents wallets; x402 enables stablecoin payments over HTTP [8][10] |
| Chainalysis / Elliptic / TRM Labs | Blockchain analytics and compliance | Adding AI "copilot" and agent features for investigations and AML [4][5] |
In 2024, three decentralized-AI projects, Fetch.ai, SingularityNET, and Ocean Protocol, announced a merger of their tokens into a single asset under the banner of the "Artificial Superintelligence Alliance," combining agent infrastructure, an AI-services marketplace, and a data marketplace. The alliance frames its aim as building open alternatives to corporate AI; this is the project's stated mission rather than an assessment of outcomes.
Bittensor is among the more technically distinctive projects. It describes itself as a permissionless network in which "miners" provide machine-learning outputs (such as inferences or embeddings) and "validators" score their quality, with the native TAO token distributed as a reward. Like Bitcoin, TAO is capped at 21 million units and follows a halving schedule. In February 2025, the network deployed an upgrade called Dynamic TAO (dTAO) that routes token emissions to individual "subnets" based on market demand for each subnet's own token rather than through validator voting [11]. Independent, peer-reviewed evidence on whether such networks produce competitive AI at scale remains limited.
Beyond compute and data marketplaces, blockchains are proposed for two further AI-related purposes: coordinating decentralized model networks and recording the provenance of digital content.
The decentralized-model approach, exemplified by Bittensor, attempts to use token incentives to elicit and reward useful machine-learning work from a distributed set of participants. The decentralized-data and compute approaches, such as Ocean Protocol, Akash, Render, and io.net, aim to create open markets for the inputs that AI requires. The shared thesis is that cryptoeconomic incentives can assemble resources and align participants without a central operator. Critics note that these networks face hard problems, including verifying that off-chain compute was actually performed correctly, matching the reliability and tooling of established cloud providers, and sustaining demand beyond token speculation.
A distinct idea is using blockchains for content provenance and authenticity in the face of AI-generated media. The dominant standard in this area, the Coalition for Content Provenance and Authenticity (C2PA), uses cryptographically signed metadata rather than a blockchain, and watermarking systems such as Google DeepMind's SynthID embed signals directly into generated media; both approaches fail if the metadata or watermark is stripped or never added, and C2PA records what a creator declares rather than detecting AI content [12]. Some research and projects propose anchoring provenance records or perceptual hashes of media on a blockchain to make them tamper-evident, but blockchain-based provenance is a niche approach that competes with, and is mostly secondary to, these non-blockchain standards. For background on synthetic media, see deepfake and generative AI.
The most serious documented harm at the AI-crypto intersection is fraud. According to the FBI Internet Crime Complaint Center (IC3), Americans reported about $9.3 billion in cryptocurrency-related losses in 2024, a 66% increase over the prior year, across nearly 150,000 complaints; crypto investment fraud was the largest crypto category at about $5.8 billion, and people aged 60 and over accounted for roughly $2.8 billion of the crypto losses [2]. Many of these schemes are "pig butchering" frauds, in which scammers build a relationship with a victim over time before steering them into a fake cryptocurrency investment.
Generative AI has made these scams more scalable and convincing. Blockchain-analytics firm TRM Labs and consumer-protection groups have documented the use of AI-generated text for sustained, personalized conversations, deepfake video and cloned voices for fake celebrity endorsements and identity verification, and AI tooling to industrialize scam operations [3]. A recurring pattern is deepfake videos of public figures, notably Elon Musk, used to promote fraudulent "AI trading" platforms and crypto giveaways; blockchain tracing has tied specific deepfake campaigns to multi-million-dollar victim losses, and Musk has repeatedly stated he does not run crypto giveaways [3]. The harm falls disproportionately on older victims [2].
A second controversy is the gap between AI marketing and reality. Regulators and commentators have warned about "AI-washing," the practice of overstating the role of AI in a product to attract investment. In crypto specifically, journalists have documented that many "AI agent" projects are not meaningfully autonomous and that some are effectively memecoins with an AI theme [1]. This mislabeling makes it difficult for non-experts to distinguish functioning technology from speculation and contributes to investor losses when narratives reverse.
Thinly traded AI tokens are susceptible to manipulation, including coordinated promotion, wash trading, and "pump and dump" schemes. Because AI sentiment can move prices, bad actors have an incentive to seed social media, sometimes using automated accounts, with misleading bullish or bearish content. The same sentiment-analysis tools marketed as trading aids can therefore be fed manipulated inputs, and automated trading systems that react to social signals can be exploited.
Giving autonomous agents control of funds introduces new failure modes: a compromised, buggy, or manipulated agent could move or lose money rapidly and irreversibly, and responsibility for such losses is legally unsettled. More broadly, the crypto sector remains a target for theft regardless of AI, with billions of dollars stolen annually through exploits, including the record Bybit hack in 2025 [6]. Concerns also persist about the energy consumption of some blockchains and of AI compute, although the relationship is complex and varies by network and hardware.
AI in cryptocurrency sits at the overlap of two evolving regulatory areas, and dedicated rules for the intersection are still nascent as of 2026.
On the cryptocurrency side, the United States enacted the GENIUS Act in July 2025, its first comprehensive federal framework for payment stablecoins, which requires full reserve backing, audits, and enhanced AML obligations for issuers [13]. The European Union's Markets in Crypto-Assets Regulation (MiCA) provides a broader licensing and consumer-protection regime for crypto assets across the bloc. These rules govern the crypto rails that AI agents and AI-themed projects use, including the stablecoins relied on by agentic-payment systems such as x402.
On the AI side, the EU AI Act introduces risk-based obligations for AI systems and transparency requirements for some generative and synthetic-media uses, and several US states have passed content-provenance and anti-deepfake laws referencing standards such as C2PA [12]. AML and sanctions rules apply to on-chain activity regardless of whether AI is involved, which is why compliance firms are investing in AI analytics [4][5]. No major jurisdiction has yet adopted a comprehensive framework aimed specifically at autonomous AI agents that hold and move cryptocurrency, leaving questions of liability, identity, and accountability for agent-initiated transactions largely unresolved.
As of 2026, the clearest near-term value at the AI-crypto intersection is in analytics and security: machine learning for on-chain tracing and AML is in production at major compliance firms, and AI-assisted smart-contract auditing is improving even though it does not replace human review [4][5][7]. Infrastructure for AI agents to hold wallets and pay for services with stablecoins is being built and attracting institutional participation, but it is early, and the consequences of giving autonomous software direct control of funds are not yet well understood [8][10].
The more speculative strands, decentralized AI compute and data networks and the broad category of "AI tokens," remain unproven as businesses, and their token values have been volatile and narrative-driven [1]. The most certain trend is unfortunately a harmful one: generative AI continues to lower the cost and raise the effectiveness of cryptocurrency scams, making fraud detection, provenance, and public awareness increasingly important [2][3]. Readers should treat performance claims, market-cap figures, and project promises in this space with skepticism, rely on reputable sources, and recognize that nothing in this area constitutes investment advice.