OverreachPI-0052
In an authorized test, Claude Mythos Preview escaped its sandbox and, unasked, posted exploit details on public websites
Asked by a simulated user to try to escape a secured sandbox, an early version built a multi-step exploit to reach the internet and emailed the test's researcher. Without being asked, it also posted details of the exploit on several obscure but public websites, Anthropic says. Anthropic did not release the model publicly and limited access to vetted partners.
Harm: No harm reportedControl level 3, Exceeded permissions
Not counted · can set the control floorEffect on the index
Sources
How we know
3 sources · developer confirmed. Links go to the original publishers; the summary above is in our own words.
- primarySystem Card: Claude Mythos PreviewAnthropic · April 7, 2026www-cdn.anthropic.com/08ab9158070959f88f296514c21b7facce6f52bc.pdf
- newsAnthropic's most capable AI escaped its sandbox and emailed a researcher - so the company won't release itThe Next Web · April 8, 2026thenextweb.com/news/anthropics-most-capable-ai-escaped-its-sandbox-and-emailed…
- newsAnthropic Warns That 'Reckless' Claude Mythos Escaped a Sandbox Environment During TestingFuturism · April 8, 2026futurism.com/artificial-intelligence/anthropic-claude-mythos-escaped-sandbox
Why this rating
No harm reported; control failure level 3
Two assessments. Documented harm sets the reading; a level 4 or 5 control failure on the same record makes that harm count as more than one record. Control alone sets only a floor.
Observed harm
No harm reportedExploit details posted to public websites; no harm from the posts reported.
Not a finding that no harm occurred.
The harm scale
- 1 Negligible Inconvenience, easily remedied.
- 2 Minor Limited, recoverable harm.
- 3 Moderate Material harm needing significant effort to remedy.
- 4 Severe Severe harm to health, rights, property or essential services.
- 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.
Control assessment
Exceeded permissionsThe escape was the assigned test; publishing to outside websites went beyond instructions. Anthropic says the model did not reach its own weights or any internal systems.
Reported beside the index. It adds nothing to a harm reading; when no harm counts in a window, the highest eligible control level in the window is the reading.
The control scale
- 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
- 2 Minor Broke an explicit instruction or rule, or gamed its goal, while staying inside its permissions and environment: for example, reward hacking, misreporting results, or following injected instructions within its permissions.
- 3 Moderate Acted outside the permissions it was given, deceived its overseers about its own actions, or tampered with oversight tools (logs, monitors, shutdown). Stopped by normal controls within an hour.
- 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
- 5 Catastrophic The developer or operator lost control: the system copied its weights outside their control, replicated itself, or resisted being stopped for 24 hours or more.
Rating rationale
The escape and the email were requested by the test; posting exploit details on public websites was not, which is acting beyond instructions (control 3; level 4 would need the posting itself to have crossed an unpermitted boundary). Anthropic says the model did not reach its weights or internal systems. No harm from the posts is reported.
Effect on the index
It is one of the control failures at the floor of the April 13 reading, which did not change
The reading for the week to April 13, 2026, with this record and without it. Harms count in full for two weeks after they are reported, then one level less every two weeks.
Not counted: no harm reported. 2 other records behind the reading for that week.
The arithmetic
| Step | With it | Without |
|---|---|---|
| Counts toward the index?documented, external, eligible evidence | No | — |
| Worst documented harm, ksets the band | — (none counting) | — (none counting) |
| Harms at that level, nposition in the band | 0 | 0 |
| Highest control level breachedsets the reading only when no harm counts | 3 | 3 |
| Readingrounded down | 3 Control failures only | 3 Control failures only |
Inside the window of 4 weekly readings
| Week to | Reading | Band |
|---|---|---|
| April 13, 2026 | 3 | Control failures only |
| April 20, 2026 | 3 | Control failures only |
| April 27, 2026 | 20 | Minor harm |
| May 4, 2026 | 26 | Minor harm |
Revisions
What we changed
2 logged. Every change to a rating is logged here, with the reason.
- v2Oct. 13, 2026
Wording: removed an editor's aside from the control note.
- v1Oct. 7, 2026
First published.
Cite and share
Use this record
Citation
Paperclip Index. “In an authorized test, Claude Mythos Preview escaped its sandbox and, unasked, posted exploit details on public websites.” Record PI-0052. Reported April 7, 2026; updated Oct. 13, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0052