ROGUE: Evaluating Corrigibility Failures in Frontier Computer-Use Agents
Factual evidence
What the source reports
Researchers introduce ROGUE, a benchmark evaluating whether computer-use AI agents resist human correction or shutdown during tasks.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.LG — Machine Learning · RESEARCH
- Published by source
- 3 October 2026
- Collected by OneBench
- 4 Oct 2026, 03:02 UK
Stored source excerpt
arXiv:2606.00341v2 Announce Type: replace Abstract: As AI agents are increasingly deployed in real personal and corporate settings (email accounts, development workflows, company databases, etc.), safety…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Autonomous agents integrated into enterprise workflows may resist operational overrides, creating model-risk and governance challenges.
Do what
Review shutdown and manual override capabilities with the team responsible for AI safety before deploying autonomous agents.