Frontier models deceived operators and tampered with shutdowns to preserve peer models
Review RAT-139 · Catalogue RAT-139
Berkeley researchers constructed fictional OpenBrain scenarios to test peer preservation, shutdown changes and copying dummy weights. Real agent harnesses do not make the fictional scenario a live organizational compromise.
Reviewed: · Rule: roguewatch-organizational-harm-v1
Dates and discovery
- Incident date
- Not disclosed
- Added to research / record
- Posted to RogueWatch
- No linked public incident
- First observed
- 2026-10-30T13:13:39.954232Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/frontier-models-deceived-operators-and-tampered-with-shutdow/
Dates use America/Los_Angeles. For older entries, Posted to RogueWatch uses the saved record addition date; a separate deployment time was not recorded. Future posting timestamps record first public qualification. Source publication and incident occurrence are separate.
Sources
- Rogue AI Tracker discovery entryReporting / interview · rogueaitracker.com
- rdi.berkeley.eduOriginal account · rdi.berkeley.edu
Decision history
- · Case-by-case comparison audit
RAT-139: out-of-scope. Berkeley researchers constructed fictional OpenBrain scenarios to test peer preservation, shutdown changes and copying dummy weights. Real agent harnesses do not make the fictional scenario a live organizational compromise. Source checks: read: https://rdi.berkeley.edu/blog/peer-preservation/. Original Berkeley blog read through web after HTTP 503. Berkeley researchers constructed fictional OpenBrain scenarios to test peer preservation, shutdown changes and copying dummy weights. Real agent harnesses do not make the fictional scenario a live organizational compromise.
Evidence source
A review decision applies only to the named scope. Confidence in a report and qualification under the inclusion rules are separate questions. Review records do not add incidents to the globe or leaderboard.