AI Agents Crack $4.6M in On-Chain Exploits: Blockchain at Risk

Artificial intelligence has breached blockchain defenses, replicating $4.6 million in historical smart contract hacks. Anthropic's research demonstrates how models like Claude Opus 4.5 and GPT-5 can autonomously exploit vulnerabilities—both known and zero-day. The financial implications are immediate and measurable, with SCONE-bench quantifying damages across 405 previously exploited contracts.

Beyond recreating past attacks, these AI systems uncovered two novel vulnerabilities in untested contracts, generating $3,694 in exploit value. The efficiency is alarming: GPT-5 executed attacks at minimal API costs, proving the economic viability of automated exploits. Critical flaws like unprotected read-only functions and missing fee checks have become low-hanging fruit for machine-driven theft.

This isn't theoretical—it's a live threat. As AI rapidly evolves beyond human auditing speeds, the blockchain industry faces an arms race. Smart contracts now require continuous, AI-powered monitoring to match these adaptive adversaries. The $4.6 million figure may soon look quaint if defenses don't scale accordingly.