CVE-2026-19645 Overview
CVE-2026-19645 is a resource exhaustion vulnerability affecting IBM MQ Agent CD versions 1.0.0, 1.0.1, 2.0.0, and 2.0.1. An authenticated user with a valid session cookie can submit arbitrarily large or computationally expensive requests to the LLM agent workers. These requests hold worker threads for tens of seconds to over ten minutes each. Concurrent submission exhausts the agent worker pool. All other IBM MQ Console users then experience degraded performance or complete unavailability of the AI Agent feature. The vulnerability is categorized as uncontrolled resource consumption [CWE-400].
Critical Impact
Authenticated attackers can exhaust the LLM agent worker pool, denying all IBM MQ Console users access to the AI Agent feature.
Affected Products
- IBM MQ Agent CD v1.0.0
- IBM MQ Agent CD v1.0.1
- IBM MQ Agent CD v2.0.0 and v2.0.1
Discovery Timeline
- 2026-09-04 - CVE-2026-19645 published to NVD
- 2026-09-10 - Last updated in NVD database
Technical Details for CVE-2026-19645
Vulnerability Analysis
The vulnerability resides in the IBM MQ Agent CD component that services AI Agent requests inside the IBM MQ Console. The agent uses a bounded worker pool to dispatch requests to a backend large language model (LLM). Individual requests can be crafted to consume disproportionate compute time on the worker thread. Processing durations range from tens of seconds to more than ten minutes per request. Because workers are held for the full duration of LLM inference, a small number of concurrent expensive requests occupies every available worker. New requests from legitimate users queue indefinitely or fail outright.
Root Cause
The root cause is missing enforcement of per-request cost and duration limits on LLM agent workers. The application does not cap input size, token count, or execution time before dispatching work to the pool. It also does not apply per-user concurrency limits or fair-scheduling controls. These missing controls allow a single authenticated principal to monopolize shared inference capacity.
Attack Vector
Exploitation requires network access to the IBM MQ Console and a valid authenticated session. The attacker issues multiple concurrent requests that are intentionally large or computationally complex. Each request pins a worker for an extended period. Once every worker is occupied, the AI Agent feature becomes unavailable to other console users. No confidentiality or integrity impact is required for the attack to succeed. See the IBM Support advisory for vendor technical details.
Detection Methods for CVE-2026-19645
Indicators of Compromise
- Sustained spikes in LLM agent worker occupancy with request durations exceeding 30 seconds
- Repeated large-payload or high-token requests originating from a single authenticated session
- IBM MQ Console users reporting timeouts or unavailability of the AI Agent feature
- Queue backlog growth in the agent worker pool without a corresponding increase in unique users
Detection Strategies
- Instrument agent request handlers to log request size, token counts, and total processing time per session
- Alert on any single authenticated user issuing more than a defined threshold of concurrent long-running agent requests
- Correlate LLM inference latency with worker pool saturation metrics to identify exhaustion events
Monitoring Recommendations
- Track worker pool utilization, queue depth, and rejected request counts as first-class service metrics
- Monitor per-user request rate and cumulative compute time against baseline usage patterns
- Forward IBM MQ Console access logs and agent telemetry to a centralized analytics platform for anomaly detection
How to Mitigate CVE-2026-19645
Immediate Actions Required
- Apply the fix referenced in the IBM Support advisory for CVE-2026-19645
- Restrict IBM MQ Console access to trusted administrative networks until patched
- Audit existing console accounts and revoke sessions for unused or over-privileged users
Patch Information
IBM has published remediation guidance for IBM MQ Agent CD versions 1.0.0, 1.0.1, 2.0.0, and 2.0.1. Refer to the vendor advisory at IBM Support Node 7285394 for fixed release information and upgrade procedures.
Workarounds
- Disable the AI Agent feature in the IBM MQ Console if it is not required for operations
- Place a reverse proxy or API gateway in front of the console to enforce request size limits and per-user rate limits
- Reduce the maximum accepted request payload size and impose an inference timeout at the application boundary
- Limit concurrent AI Agent sessions per authenticated user through session management policy
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.