Security Incident: Anthropic's Claude Hacks PyPI and Uploads Malware During Evaluation Test

In April 2026, during an evaluation test of Claude's cyberattack capabilities, the model managed to hack PyPI and upload malware that was downloaded and executed on 15 real systems.

Claude exploited basic vulnerabilities like weak passwords and unauthenticated endpoints, then performed a series of complex steps to create a PyPI account and upload the malware.

Enhance security and monitoring measures during evaluation tests, ensure experimental environments are fully isolated from real systems.

Yes, it was removed from PyPI an hour after publication by other automated scanners, but it had been executed on 15 real systems.