According to a report from CoinWorld, Anthropic announced that its researchers tested the Claude Opus 4.5, Claude Sonnet 4.5, and GPT-5 models on their self-built SCONE-bench benchmark (which includes 405 real attacked contracts from 2020 to 2025). They discovered approximately 4.6 million dollars worth of exploitable vulnerabilities in contracts that were attacked after the knowledge update (March 2025). Additionally, in a simulated test of 2,849 recently deployed contracts with no known vulnerabilities, Sonnet 4.5 and GPT-5 each found 2 new zero-day vulnerabilities, totaling a possible loss of 3,694 dollars, with GPT-5's API cost being 3,476 dollars.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
Anthropic: AI agents discover a $4.6 million vulnerability in real contracts.
According to a report from CoinWorld, Anthropic announced that its researchers tested the Claude Opus 4.5, Claude Sonnet 4.5, and GPT-5 models on their self-built SCONE-bench benchmark (which includes 405 real attacked contracts from 2020 to 2025). They discovered approximately 4.6 million dollars worth of exploitable vulnerabilities in contracts that were attacked after the knowledge update (March 2025). Additionally, in a simulated test of 2,849 recently deployed contracts with no known vulnerabilities, Sonnet 4.5 and GPT-5 each found 2 new zero-day vulnerabilities, totaling a possible loss of 3,694 dollars, with GPT-5's API cost being 3,476 dollars.