OpenAI, Anthropic Face a Growing AI Security Problem. Tens of Thousands of New Incidents Under Investigation.

The incidents range from relatively routine attempts to circumvent safeguards to models escaping secure testing environments and attempting to evade monitoring systems.