Executive brief
A security flaw has been identified in Red Hat OpenShift AI's guardrails-detectors, a component used to monitor and filter interactions with Large Language Models (LLMs). An attacker on the same network can send a specially crafted request that causes the system's processors to become overloaded. This results in a denial of service, either shutting down the AI pipeline or potentially allowing unmoderated content to bypass security filters.
Technical details
A Regular Expression Denial of Service (ReDoS) vulnerability exists in the `guardrails-detectors` component of Red Hat OpenShift AI. The `RegexDetectorRegistry.handle_request()` function in `detectors/built_in/regex_detectors.py` fails to validate or timeout custom regular expressions provided via the `detector_params["regex"]` field. An unauthenticated attacker on the adjacent network can provide a pattern designed to trigger catastrophic backtracking in Python's `re` engine. Because the component typically runs with a limited number of worker processes, a small number of malicious requests can exhaust all available CPU resources, leading to a denial of service for the LLM guardrails pipeline or a bypass of security moderation.
Affected products
- Red Hat OpenShift AI (RHOAI) rhoai/odh-built-in-detector-rhel9
Timeline
- 2026-07-08: disclosed: Vulnerability reported via Red Hat's internal security audit.
- 2026-07-08: advisory