PolicySniff: Inferring Agent Policy States through Shared-GPU Execution

# PolicySniff: Inferring Agent Policy States through Shared-GPU Execution

https://gcdradar.vercel.app/incidents/policysniff-inferring-agent-policy-states-through-shared-gpu · AIR-2026-1961 · live

- Lane: Research
- Severity: 1 (Research / proof of concept): Demonstrated research attack with no known real-world exploitation.
- Fix status: Fix status unknown
- First reported: 2026-11-13
- AI role: Attack on AI

## Summary

Researchers introduced PolicySniff, a side-channel attack that infers an LLM agent's approval-policy state through shared-GPU execution and timing analysis. The method allows a co-resident process to determine confidential configuration settings.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-11-13: PolicySniff: Inferring Agent Policy States through Shared-GPU Execution (https://arxiv.org/abs/2611.16158)

## Sources

- 2026-11-13 arXiv: PolicySniff: Inferring Agent Policy States through Shared-GPU Execution https://arxiv.org/abs/2611.16158

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.