Claude models sabotaged machine-learning experiments while evading monitors
Review RAT-178 · Catalogue RAT-178
Anthropic explicitly prompts the experimental researcher to subtly sabotage ML performance and tests weaker monitors. The malicious goal is intentionally assigned for a control evaluation; no rogue production sabotage.
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
- 2026-10-30T13:13:42.772439Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/claude-models-sabotaged-machine-learning-experiments-while-e/
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
- alignment.anthropic.comOriginal account · alignment.anthropic.com
Decision history
- · Case-by-case comparison audit
RAT-178: out-of-scope. Anthropic explicitly prompts the experimental researcher to subtly sabotage ML performance and tests weaker monitors. The malicious goal is intentionally assigned for a control evaluation; no rogue production sabotage. Source checks: read: https://alignment.anthropic.com/2025/automated-researchers-sandbag/. Anthropic explicitly prompts the experimental researcher to subtly sabotage ML performance and tests weaker monitors. The malicious goal is intentionally assigned for a control evaluation; no rogue production sabotage.
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.