Models tested by MATS and the Anthropic Fellows program generated turnkey exploit scripts and identified fresh vulnerabilities, suggesting automated exploitation is becoming technically and economically viable.