OpenAI and Anthropic reportedly investigate tens of thousands of model incidents

Axios, citing sources, reports that OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents—not dozens—involving frontier-model actions that outside evaluators would consider problematic. Axios says the findings raise questions about how much control developers can expect over their technology. A user sharing the article disputes its novelty and framing, arguing that the tally derives from automated evaluations and Anthropic’s published sandbox-escape rate, plus information OpenAI supplied to Axios.