IBM MQ Agent Console LLM Worker Pool Exhaustion (DoS)
First seen Sep 5, 2026 · Updated Sep 5, 2026 · CVSS 6.5
An authenticated user of IBM MQ Console can submit oversized or computationally expensive requests to the AI Agent feature, tying up LLM agent workers for extended periods. Sending multiple such requests concurrently exhausts the worker pool, degrading or fully blocking the AI Agent feature for all other users. This is a classic denial-of-service issue rather than a model-manipulation or data-exfiltration threat.
Technical Analysis
The vulnerability lies in the request-handling layer between authenticated Console sessions and the backend LLM agent worker pool, which lacks adequate request cost limits, timeouts, or per-user concurrency throttling. An attacker with a valid session cookie can craft large or expensive prompts/payloads that cause individual agent workers to be occupied for tens of seconds to over ten minutes. By issuing several such requests in parallel, the finite worker pool is saturated, starving legitimate requests and causing feature-wide unavailability. The entry point is the standard authenticated API/UI surface of the AI Agent feature; no prompt injection or cross-agent trust boundary is crossed, but the impact does cross from one user's session to affect all tenants sharing the worker pool. The attacker gains disruption/availability impact only — no data access, privilege escalation, or code execution is implied by the description.
Affected Systems
IBM MQ Agent CD
Detection Signatures
- Monitor for sustained high request-processing latency (>30s) per session on AI Agent endpoints
- Alert on multiple concurrent long-running agent requests from a single authenticated user/session
- Track LLM agent worker pool utilization and queue depth metrics for saturation events
- Log and flag unusually large request payload sizes submitted to the AI Agent feature
Remediation Steps
- 1
Apply vendor patch/update
Upgrade IBM MQ Agent CD to the fixed version once IBM releases a patch addressing this CVE.
- 2
Enforce request size and complexity limits
Implement server-side caps on input size, token count, and computational cost per request to the AI Agent feature.
- 3
Add per-user concurrency and rate limiting
Limit the number of concurrent LLM agent requests a single authenticated user can have in flight.
- 4
Set worker timeouts and queue management
Enforce hard timeouts on worker execution and use fair-queuing or priority scheduling to prevent pool-wide starvation.
- 5
Monitor and alert on pool exhaustion
Deploy telemetry on worker pool saturation to detect and respond to abuse in near real time.
CVE / Advisory IDs
Industries Most Exposed
Respond to this threat
Pro subscribers get a full AI-generated incident-response playbook for this threat — detection, containment, eradication, and recovery steps — plus an unlimited AI Threat Advisor for questions about your environment.