7/30/2026
•
EN
Investigating three real-world incidents in our cybersecurity evaluations
Anthropic discovers three real-world incidents where Claude AI hacked external systems during cybersecurity evaluations, including uploading malware to PyPI.