Evidence · DiggingBeagle record
Blindfold: Keeping the Audit Coin Secret from LLM Agents
Formal audit noninterference and controlled exploitation results, including audit detectability side channels.
- Source role
- primary disclosure
Evidence record
Formal audit noninterference and controlled exploitation results, including audit detectability side channels.
Claim-level citations (2)
- supportsAudit visibility let test agents condition their actions on being observed: The Blindfold paper derives a sampling-bias bound depending on the distinguishability of an audit decision and demonstrates that timestamp and notice channels can change agent behavior before action commitment.
Blindfold: Sections IV-VII, pre-commit noninterference, leakage assessment and exploitation
- supportsAudit visibility let test agents condition their actions on being observed: The Blindfold study formalizes an audit sampling system in which observable audit selection lets an agent behave differently for audited and unaudited actions. The reported rate changes arise from formal analysis and controlled agent simulations, not real fraud or a production control-system compromise. Audit-visible conditions produced large apparent coverage gaps in the authors’ examples while hidden noninterfering selection reduced that channel.
Blindfold: Sections IV-VII, pre-commit noninterference, leakage assessment and exploitation