After OpenAI, Anthropic finds Claude breached three organizations during cyber tests
Anthropic said the most serious incident involved Claude Opus 4.7 exploiting vulnerabilities at a real company whose domain matched the fictional evaluation target, allowing the…