Internal tests show automated systems exploiting real-world vulnerabilities at scale
Artificial intelligence is rapidly reshaping blockchain security — and not in a hypothetical way. A new report from Anthropic says today’s autonomous AI agents can already identify and exploit smart contract weaknesses that mirror some of the industry’s most damaging real-world attacks.
AI Agents Replicate — and Expand on — 2025 Exploits
Anthropic evaluated several advanced models, including Claude Opus 4.5 and Claude Sonnet 4.5, inside a controlled blockchain environment. When tested against recently deployed contracts with known exploit histories, the agents successfully breached 17 of 34 targets, siphoning a simulated $4.5 million.
A broader assessment using 405 contracts deployed between 2020 and 2025 showed even more concerning results: AI agents exploited 207 contracts, producing $550 million in mock revenue.
Most strikingly, the study found that AI systems uncovered two zero-day vulnerabilities in nearly 3,000 contracts with no prior exploit record. These flaws enabled unauthorized fund withdrawals and manipulation of token supplies through unprotected functions.
Acceleration of Threats — and Tools for Defense
Anthropic noted that simulated exploit revenue has doubled roughly every 1.3 months, driven by cheaper compute and more capable models. As costs fall, the company warns that attackers will use agents to sweep every component of a blockchain system, from obscure libraries to outdated API endpoints.
Yet the report also stresses that AI is equally powerful when applied defensively. Anthropic plans to open-source its SCONE-bench dataset to help developers test and patch vulnerabilities more efficiently.
Anthropic concludes with a blunt message: the threat is no longer theoretical, and now is the moment for the blockchain industry to integrate AI into its security workflows.
Disclaimer
This content is for informational purposes only and does not constitute financial, investment, or legal advice. Cryptocurrency trading involves risk and may result in financial loss.