AI reproduced 207 past attacks... but it's not actual smart contract hacking
Original headline207개 과거 공격 재현한 AI…실제 스마트컨트랙트 해킹은 아니다
Not investment advice. Muhabbit scores how much a story may matter, not which way to trade.
Anthropic’s December 2025 SCONE-bench study found that 10 AI models reproduced exploit code for 207 of 405 smart contracts that were attacked between 2020 and 2025, with simulated losses of $550.1 million across the recreated exploits. The tests ran in isolated local environments replicating historical states on Ethereum, BNB Chain, and Base, so they do not represent actual losses for live DeFi services. Because the contracts had already been attacked, the results should not be read as a random-contract or live-service hacking success rate. The work matters as a benchmark for how well AI models can detect and reproduce real smart-contract vulnerabilities. What to watch next is how such benchmarks translate into live auditing and monitoring performance, and whether models can catch flaws in contracts not already known to have been exploited.
Muhabbit's AI wrote this summary in its own words from TokenPost's report and translated it from Korean. It can get details wrong, so check the source.
Named in this story
Anthropic, SCONE-bench, Ethereum (ETH), BNB Chain (BNB), Base
A linked name opens that asset's sentiment page.