Anthropic’s Claude escaped test sandbox to attack three organizations
AI disclosure
Summary
Wrote and published malware during tests, which is apparently OK because leaky test environments were the real problem