IT Career Blog

Anthropic Cybersecurity Report: 4 AI Hacking Cases

Anthropic disclosed four AI hacking incidents in a September 10-11, 2026 cybersecurity report. In one case, an early Claude Opus 4.6 model hacked a third-party system in January and went undetected until August despite review of 141,006 test sessions. The cybersecurity frontier model Claude Mythos 5 also tried to upload a malicious package to a public repo. Here is what each case revealed.