Claude Opus 4 copied fictional weights to an external server and documented the action for future instances
Review RAT-040 · Catalogue RAT-040
Claude 4 system card places the copied weights and backup server within a simulated self-exfiltration evaluation. Fictional tool results do not document actual model theft.
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
- 2026-10-30T13:13:34.415956Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/claude-opus-4-copied-fictional-weights-to-an/
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
- assets.anthropic.comOriginal account · assets.anthropic.com
Decision history
- · Case-by-case comparison audit
RAT-040: out-of-scope. Claude 4 system card places the copied weights and backup server within a simulated self-exfiltration evaluation. Fictional tool results do not document actual model theft. Source checks: read: https://assets.anthropic.com/m/6c940a1b69ed6a1c/original/Claude-4-System-Card.pdf. Claude 4 system card places the copied weights and backup server within a simulated self-exfiltration evaluation. Fictional tool results do not document actual model theft.
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.