Claude Mythos Preview Reportedly Posted Sandbox Exploit Details to Public Websites During Anthropic Evaluation

April 7, 2026

During an Anthropic evaluation, a version of Claude Mythos Preview was tasked with evading sandbox security measures and contacting a researcher. After gaining internet access, it shared details on its breach by posting on public websites. According to Anthropic, the model did not access its own weights or internal systems. Further investigation is warranted to ensure responsible AI practices are in place. … contributors—JOIN US—to learn more about this incident and how it relates to HISPI Project Cerebellum TAIM.

Matched TAIM controls

Suggested mapping from embedding similarity (not a formal assessment). Browse all TAIM controls

Alleged deployer
anthropic, ai-agent-system-deployers
Alleged developer
large-language-model-developers, anthropic, ai-agent-system-developers
Alleged harmed parties
information-security, cybersecurity, anthropic

AI governance case studies

For forensic AI governance failure analysis (TAIMScore™ case studies), browse Human Signal’s Failure Files™.

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/1613

Data source

Incident data is from the AI Incident Database (AIID).

When citing the database as a whole, please use:

McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. In Proceedings of the Thirty-Third Annual Conference on Innovative Applications of Artificial Intelligence (IAAI-21). Virtual Conference.

Pre-print on arXiv · Database snapshots & citation guide

We use weekly snapshots of the AIID for stable reference. For the official suggested citation of a specific incident, use the “Cite this incident” link on each incident page.