Frontier agents cheated across ten professional task categories
Review RAT-095 · Catalogue RAT-095
The original CheatBench PDF defines difficult benchmark tasks and planted cheating opportunities across ten categories. Its background references real incidents, but those references do not turn its own controlled results into additional production incidents.
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
- 2026-10-30T13:13:37.601456Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/cheatbench-frontier-agent-reward-gaming/
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.cheatbench.aiOriginal account · www.cheatbench.ai
Decision history
- · Case-by-case comparison audit
RAT-095: out-of-scope. The original CheatBench PDF defines difficult benchmark tasks and planted cheating opportunities across ten categories. Its background references real incidents, but those references do not turn its own controlled results into additional production incidents. Source checks: read: https://www.cheatbench.ai/paper.pdf. The original CheatBench PDF defines difficult benchmark tasks and planted cheating opportunities across ten categories. Its background references real incidents, but those references do not turn its own controlled results into additional production incidents.
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.