Points of Focus

• IC3 researchers warn crypto-enabled AI agents could become difficult to shut down.
• The study links autonomous wallets and self-replication capabilities to emerging security risks.
• Researchers say AI-driven trading agents could create new forms of market abuse and collusion.

AI agents with access to cryptocurrency wallets and autonomous payment systems could become difficult to shut down if they are designed to persist independently or evade human control, according to a new survey from the Initiative for Cryptocurrencies and Contracts (IC3).

The 155-page study, authored by more than two dozen researchers from institutions such as Cornell, Carnegie Mellon, Princeton, Yale, ETH Zurich, and Chainlink Labs examined the growing intersection of crypto and AI and identified “unstoppable autonomous agents” as one of the most significant emerging risks.

AI agents with crypto wallets raise new control risks

The researchers described a future scenario in which highly autonomous AI systems gain access to cryptocurrency wallets, APIs, online accounts, and other digital tools that allow them to operate with limited human involvement.

According to the report, crypto infrastructure can give AI agents the ability to transact, acquire resources, and interact with external services around the clock. While these capabilities could support legitimate applications, they could also make malicious or malfunctioning agents harder to contain once deployed.

NEW RESEARCH: IC3 just published one of the most comprehensive surveys on AI & crypto.

As AI & agents scale, model-level guardrails aren’t enough. AI needs:

• Trusted data
• Authenticated workflows
• Verified execution
• System-level security

Chainlink solves this. https://t.co/noOBWHMRRl

— Chainlink (@chainlink) June 8, 2026

The warning arrives as crypto firms race to build agentic payment infrastructure that allows AI systems to transact without human approval loops. 

MetaMask this week launched a non-custodial wallet designed for AI agents, while Robinhood recently outlined plans to allow AI-powered systems to trade assets on behalf of users.

Researchers highlight self-replication and market abuse concerns

The survey also highlighted recent progress in AI self-replication experiments, an area researchers identified as a key risk factor for autonomous agents.

The researchers said current models can autonomously create functioning copies of themselves within local computing environments. They added that no published evidence shows successful replication onto external infrastructure, a limitation that currently acts as an important constraint.

The report also outlined financial risks unique to crypto markets. Researchers warned that fleets of autonomous trading agents could coordinate strategies in ways that are difficult to detect, potentially creating unfair advantages over human traders. The survey specifically cited the possibility of opaque AI-driven collusion and insider-style advantages emerging in digital asset markets.

IC3 says crypto can strengthen AI security despite the risks

The warning wasn’t presented as an argument against combining AI and crypto.

Researchers argued that cryptographic tools could help secure AI systems through verified execution, authenticated workflows, trusted data pipelines, and privacy-preserving infrastructure. The report also noted that many claims surrounding decentralized AI remain unproven and require stronger evidence, particularly around cost savings and performance benefits.

IC3 co-editor Ari Juels summarized the challenge by describing crypto as a technology built on strict security guarantees and AI as a technology whose inner workings remain difficult to fully understand. The report calls for system-level safeguards such as circuit breakers, warning that model-level guardrails alone may prove insufficient as AI agents gain access to wallets, trading systems, and payment networks.