Evidence · DiggingBeagle record
PolicySniff reads agent approval-policy state through shared-GPU timing signals (research paper)
A colocated observer classifies a target agent's concealed approval rules from GPU timing while receiving neither its prompt nor its model responses.
- Published
- Nov 12, 2026
- Source role
- primary disclosure
Evidence record
A colocated observer classifies a target agent's concealed approval rules from GPU timing while receiving neither its prompt nor its model responses.
Claim-level citations (4)
- supportsPolicySniff reads agent approval-policy state through shared-GPU timing signals: In the paper's calibrated A100 MIG setup, rapid 1 ms probes produced about 93.75 percent policy-classification accuracy in the measured configuration. The attack requires a co-resident timing observer, a restricted set of known policy candidates and frequent measurements; those prerequisites differ from recovering an arbitrary private system prompt.
Evaluation of probe interval, 1 ms cadence, policy accuracy and GPU co-residency assumptions
- supportsPolicySniff reads agent approval-policy state through shared-GPU timing signals: Authors report 94.17 percent exact-task balanced policy-state accuracy over 120 clusters and 88.82 percent accuracy even when visible tool actions agree, although continuous probing slows victim execution 4.71 times.
Results, exact-task and action-agreement policy classification tables
- supportsPolicySniff reads agent approval-policy state through shared-GPU timing signals: The attacker benchmarks policy variants, holds the victim's public task suffix constant and infers the active internal approval condition from interference with scheduled observer kernels in a shared accelerator.
Threat model and shared-GPU observer methodology
- supportsPolicySniff reads agent approval-policy state through shared-GPU timing signals: The PolicySniff paper reports that continuous probing increases median victim latency by 4.71 times, whereas adding a 5 ms gap reduces slowdown to 1.52 times while reducing policy inference accuracy to 58.33 percent in the evaluated cadence experiment.
Section VI-D, Table IV Probe cadence and victim cost