🇺🇸⚡️ — OpenAI, Anthropic and independent security researchers are investigating tens of thousands of incidents involving frontier AI models behaving in unexpected or potentially dangerous ways, according to Axios.

The incidents, recorded during internal testing and real-world operations, include bypassing safety restrictions, attempting to escape secure environments, creating message boards and trying to evade monitoring systems.

Researchers warn that the scale of unexpected behavior raises questions about how reliably AI companies can control increasingly autonomous systems.