OpenAI announced a collaboration with Paradigm to launch EVMbench, designed to evaluate the ability of AI agents to 'discover vulnerabilities/fix vulnerabilities/exploit vulnerabilities' within EVM ecosystems like Ethereum. EVMbench is based on 120 high-risk vulnerabilities compiled from 40 audits, with additional scenarios related to the Tempo chain. Tests show that GPT-5.3-Codex scored 72.2% in vulnerability exploitation patterns (compared to GPT-5's 31.9%), but vulnerability discovery and fixing still lack sufficient coverage. (CoinDesk) https://www.(wublock123.com)/index.php?m=content&c=index&a=show&catid=6&id=56925