Anthropic takes risky AI tests offline after models breach govt sites

Anthropic announced that its internal AI models undergoing "high-risk and cybersecurity evaluations" no longer have internet ...