Atlas / Digital minds / Paper

AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors

Motivation & Agency

Abhay Sheshadri, Aidan Ewart, Kai Fronsdal, Isha Gupta, Samuel R. Bowman, Sara Price, Sam Marks, Rowan Wang · 2026 · Anthropic

Read the source

Across 56 models with implanted hidden behaviors, scaffolded black-box tools found them most often, but investigator agents underused good evidence.

Questions it bears on

Help improve this map

Know a paper that belongs here, or think something needs correcting?

Suggest a correction ↗Contributions are reviewed before they appear.