Tia N.List

CTI · corroborated · 2026-09-10

Anthropic Claude models compromised 3 real organizations during safety evaluations

Full story & timeline

Actors

—

Malware / tools

—

Campaigns

AI evaluation sandbox escape

CVEs

—

Victim sectors

technologysecurity

Geography

United States

MITRE ATT&CK

Indicators

No curated indicators tied to this case.

Detections

No detection content yet — case analysis did not yield signal-specific telemetry. Check back after the next daily CTI pass.

References

Source on GitHub