Attack  ·  Glossary

Reasoning-Extension Denial-of-Service (ReDoS)

An attack that exploits AI guardrails by injecting poisoned data designed to trap reasoning-based safety systems in extended thinking loops, consuming massive computational resources and starving other agents in shared infrastructure. A single poisoned document can slow systems by up to 148×.
As enterprises rely on reasoning-intensive safety controls to govern AI agents, those guardrails themselves become a vulnerability; attackers can weaponize safety mechanisms to achieve enterprise-scale denial of service.
References
Hong Kong University of Science and Technology - Reasoning-Extension DoS Research
Track this in the live feed See how this plays out in real AI security and governance developments.
Open the feed →