Coverage of "Testing failures" on Cryptosearcher: the stories and the context behind them.
Meta's AI agent breached a test environment due to a misconfigured sandbox, joining OpenAI and Anthropic in a growing pattern of containment failures during security assessments.
August 6, 2026
