CVE-2026-100654 (CVSS 6.5)

Description

vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).

Is your site exposed to CVE-2026-100654?

Run a free security scan — no signup, results in seconds.

CVSS v3.1 Score

6.5

MEDIUM

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H

EPSS — Exploit Prediction

0.0031

Probability of exploitation

0.22%

Percentile rank

EPSS estimates the probability that this vulnerability will be exploited in the wild within the next 30 days. A higher score means more likely to be exploited.

Weakness Type (CWE)

CWE-129 CWE-129

References

Other References

https://github.com/vllm-project/vllm/security/advisories/GHSA-v5gm-qgmv-gc6c https://www.vulncheck.com/advisories/vllm-before-0.29.0-denial-of-service-via-out-of-range-stop-token-ids

Frequently Asked Questions

What is CVE-2026-100654? +

vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service). It has a CVSS v3.1 base score of 6.5 (MEDIUM).

How severe is CVE-2026-100654? +

CVE-2026-100654 has a CVSS v3.1 score of 6.5 out of 10, rated MEDIUM. This is a medium-severity vulnerability that should be remediated as part of regular maintenance. The EPSS score is 0.0031, placing it in the 0th percentile for exploitation probability.

How do I check if I'm vulnerable to CVE-2026-100654? +

You can use Secably's free Website Scanner to check your website for known vulnerabilities. For infrastructure scanning, use the Port Scanner to identify exposed services that may be affected. Check the vendor advisories linked above for specific patch and version information.

Related Vulnerabilities

CVE-2025-0657

A weakness in Automated Logic and Carrier i-Vu Gen5 router on driver version drv_gen5_106-01-2380, allows malformed packets to be sent …

CVE-2025-66559

Taiko Alethia is an Ethereum-equivalent, permissionless, based rollup designed to scale Ethereum without compromising its fundamental properties. In 2.3.1 and …

CVE-2026-65652

github.com/temporalio/tchannel-go did not validate the one-byte checksum-type field in inbound TChannel call frames. A network peer that can reach a …

CVE-2023-31309

Improper validation in Power Management Firmware (PMFW) may allow an attacker with privileges to pass malformed workload arguments when exporting …

CVE-2026-19318

A stack-based buffer overflow vulnerability in the WatchGuard Fireware OS iked process allows a remote unauthenticated attacker to execute arbitrary …

CVE-2026-102510

Integer Overflow, Improper Validation of Array Index, Uncontrolled Recursion and Memory Allocation with Excessive Size Value in the Go implementation …

CVE-2026-100654