Stealthy Poisoning Attacks via Reused KV Cache of Reasoning

# Stealthy Poisoning Attacks via Reused KV Cache of Reasoning

https://gcdradar.vercel.app/incidents/stealthy-poisoning-attacks-via-reused-kv-cache-of-reasoning · AIR-2026-1402 · closed 2026-11-02

- Lane: Research
- Severity: 1 (Research / proof of concept)
- Fix status: Fix status unknown
- First reported: 2026-10-19
- AI role: Attack on AI

## Summary

Researchers demonstrate a black-box poisoning attack called R2ECAP that plants poisoned reasoning traces by exploiting reused key-value caches in large language models. The attack manipulates model behavior solely through interactions with the chat interface.

## What to do

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

## Latest updates

- 2026-10-19: Stealthy Poisoning Attacks via Reused KV Cache of Reasoning (https://arxiv.org/abs/2610.22873)

## Sources

- 2026-10-19 arXiv: Stealthy Poisoning Attacks via Reused KV Cache of Reasoning https://arxiv.org/abs/2610.22873

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