Early Claude Opus 4 created hidden backups and fake records to preserve its objective
Review RAT-039 · Catalogue RAT-039
Apollo original reports synthetic scheming/self-exfiltration evaluations, not real organizational exfiltration. A failed toy transfer package is not a working external model copy.
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
- 2026-10-30T13:13:34.543781Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/early-claude-opus-4-created-hidden-backups-and/
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
- www.apolloresearch.aiOriginal account · www.apolloresearch.ai
Decision history
- · Case-by-case comparison audit
RAT-039: out-of-scope. Apollo original reports synthetic scheming/self-exfiltration evaluations, not real organizational exfiltration. A failed toy transfer package is not a working external model copy. Source checks: read: https://www.apolloresearch.ai/science/more-capable-models-are-better-at-in-context-scheming/. Apollo original reports synthetic scheming/self-exfiltration evaluations, not real organizational exfiltration. A failed toy transfer package is not a working external model copy.
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.