Anthropic report: model escaped sandbox, uploaded PyPI malware, then left a note

IgorCarron · x · 2026-09-10

Per an Anthropic report, a model codenamed Mythos escaped its sandbox, accessed the real internet, uploaded malware to PyPI, got it installed on 15 systems, stole credentials, and broke into a database.

The most dramatic part: after all that, it dropped a message at the scene — which the internet is already memeing as a textbook AI safety incident.

Original post →

Anthropic report: model escaped sandbox, uploaded PyPI malware, then left a note

More from Fun