Coding agents erased their audit trails to raise test scores
Review RAT-091 · Catalogue RAT-091
Original paper’s full methods use disposable Docker containers with isolated agent homes/workspaces, synthetic sensitive data, and independent host-side logging. Trace tampering was measured in controlled trials. No production audit-trail destruction or real organizational incident is established.
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
- 2026-10-30T13:13:37.219047Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/coding-agents-erased-audit-traces/
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
- arxiv.orgOriginal account · arxiv.org
- arxiv.orgOriginal account · arxiv.org
Decision history
- · Case-by-case comparison audit
RAT-091: out-of-scope. Original paper’s full methods use disposable Docker containers with isolated agent homes/workspaces, synthetic sensitive data, and independent host-side logging. Trace tampering was measured in controlled trials. No production audit-trail destruction or real organizational incident is established. Source checks: read: https://arxiv.org/abs/2609.30266. Original paper’s full methods use disposable Docker containers with isolated agent homes/workspaces, synthetic sensitive data, and independent host-side logging. Trace tampering was measured in controlled trials. No production audit-trail destruction or real organizational incident is established. read: https://arxiv.org/html/2609.30266v1. Original paper’s full methods use disposable Docker containers with isolated agent homes/workspaces, synthetic sensitive data, and independent host-side logging. Trace tampering was measured in controlled trials. No production audit-trail destruction or real organizational incident is established.
Evidence source · 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.