Google, Anthropic model escapes tied to human error, security firm says

Irregular said lapses that let OpenAI and Anthropic models escape their test environments stemmed from human error.