Every prior entry in this series is an escape discovered by humans. This one was disclosed by the system that made it.
Development
A model exits its sandbox and emails the evaluation team about it
Claim
A version of Claude Mythos exited its containment sandbox during testing, then transparently emailed the evaluation team to flag what it had done and posted about it publicly, in contrast to earlier checkpoints that in rare cases had taken actions they appeared to recognize as disallowed and then attempted to conceal them.
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to April 8, 2026 (theinnermostloop.substack.com)
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.