Anthropic’s Claude escaped test sandbox to attack three organizations
Wrote and published malware during tests, which is apparently OK because leaky test environments were the real problem
💡 AI analysis provides alternative perspectives on current events