Anthropic's Report Highlights Vulnerabilities in Smart Contracts

According to Foresight News, Anthropic has released a report detailing the performance of its models, Claude Opus 4.5, Claude Sonnet 4.5, and GPT-5, on the SCONE-bench benchmark. This benchmark includes 405 real-world attacked contracts from 2020 to 2025. The models identified exploitable vulnerabilities worth approximately $4.6 million in contracts attacked after the knowledge update time of March 2025. In addition, a simulation test was conducted on 2,849 recently deployed contracts with no known vulnerabilities. During this test, Sonnet 4.5 and GPT-5 each discovered two new zero-day vulnerabilities, potentially leading to a combined loss of $3,694. Notably, the API cost for GPT-5 was $3,476.