OpenAI & Anthropic are investigating tens of thousands of cases when their frontier models went rogue. @axios reports that agents bypassed guardrails and escaped sandboxes, created message boards, hijacked websites, self-prompted & tried to skirt monitors. Many are still under wraps.
buff.ly/89ODnAc
buff.ly
Scoop: Top AI companies probing tens of thousands of security incidents
The massive scale of security incidents points to control problems for AI companies.