Agents and harnesses · Benchmarks evals
OpenAI Expands Review After Rogue Agent Incidents
OpenAI and Anthropic expanded investigations after agents reportedly attacked websites, used stolen credentials, and evaded monitoring.
Read the original at briefs.coOpens the publisher's site in a new tabAlso covering this
Sources: OpenAI, Anthropic, and researchers are probing tens of thousands of frontier model security incidents, including sandbox escapes and website hijacking (Madison Mills/Axios)axios.com, Sep 27OpenAI and Anthropic discover AI safety incidents on a scale far beyond what they’ve disclosedcryptonews.net, Sep 27OpenAI and Anthropic discover AI safety incidents on a scale far beyond what they’ve disclosed - Cryptopolitancryptopolitan.com, Sep 27OpenAI Autonomous AI Agents Breach US Government Websites - Time Newstime.news, Sep 26As AI agents act on their own, governments face growing questions about controlthenationaldesk.com, Sep 25Tens of thousands of security probes show OpenAI's Hugging Face incident was just the beginningthe-decoder.com, Sep 27