CVE-2026-103241: vLLM: Gemma4 parser flaw allows remote DoS
HIGHPoC AVAILABLECISA: TRACK*A flaw in the Gemma4UnifiedParser component of vLLM (versions up to 0.26.0) lets an unauthenticated remote attacker crash or exhaust the inference service with a crafted input, causing denial of service (CWE-404, improper resource shutdown or release). It scores CVSS 7.5 (network, low complexity, no privileges, no user interaction), and a public PoC and exploit already exist (a gist plus a GitHub issue). Even so, the EPSS score of 0.00714 (top 48%) and the absence from CISA KEV suggest no mass exploitation yet. The CISA SSVC decision is TRACK_STAR, so it should be monitored closely. Any internet-exposed vLLM endpoint serving Gemma-family models is a cheap target for knocking out AI availability. Upgrade to 0.29.1rc0 or later, which includes commit 3439bad, and restrict access to the inference API until that is done.
What is the risk?
Medium-high. Technical severity is high (CVSS 7.5, availability impact only). Exploitation is trivial given the public PoC and the lack of authentication. Real exposure depends on whether the vLLM endpoint is reachable by untrusted clients and whether the Gemma4 unified parser is in use. EPSS is low (top 48%), there is no KEV listing and no Nuclei template, so there is no evidence of mass exploitation. There is no confidentiality or integrity impact.
How does the attack unfold?
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| vLLM | pip | — | No patch |
Do you use vLLM? You're affected.
How severe is it?
What is the attack surface?
What should I do?
1 step1) Upgrade vLLM to 0.29.1rc0 or later (patch 3439bad37e68ba9755a46f4f6b44a4aeaf1f60a9, PR #54303). 2) Until you can patch, place the vLLM API behind an authenticated gateway or reverse proxy with rate limiting and request size limits, and do not expose it directly to the internet. 3) If it is not needed, avoid serving Gemma4 models with the unified parser. 4) Detection: watch for abnormal restarts or OOM kills of vLLM processes, latency spikes, and repeated malformed requests from the same source. 5) Configure automatic restarts and health checks to limit downtime, and review issue #50927 for technical details.
What does CISA's SSVC say?
Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is CVE-2026-103241?
A flaw in the Gemma4UnifiedParser component of vLLM (versions up to 0.26.0) lets an unauthenticated remote attacker crash or exhaust the inference service with a crafted input, causing denial of service (CWE-404, improper resource shutdown or release). It scores CVSS 7.5 (network, low complexity, no privileges, no user interaction), and a public PoC and exploit already exist (a gist plus a GitHub issue). Even so, the EPSS score of 0.00714 (top 48%) and the absence from CISA KEV suggest no mass exploitation yet. The CISA SSVC decision is TRACK_STAR, so it should be monitored closely. Any internet-exposed vLLM endpoint serving Gemma-family models is a cheap target for knocking out AI availability. Upgrade to 0.29.1rc0 or later, which includes commit 3439bad, and restrict access to the inference API until that is done.
Is CVE-2026-103241 actively exploited?
Proof-of-concept exploit code is publicly available for CVE-2026-103241, increasing the risk of exploitation.
How to fix CVE-2026-103241?
1) Upgrade vLLM to 0.29.1rc0 or later (patch 3439bad37e68ba9755a46f4f6b44a4aeaf1f60a9, PR #54303). 2) Until you can patch, place the vLLM API behind an authenticated gateway or reverse proxy with rate limiting and request size limits, and do not expose it directly to the internet. 3) If it is not needed, avoid serving Gemma4 models with the unified parser. 4) Detection: watch for abnormal restarts or OOM kills of vLLM processes, latency spikes, and repeated malformed requests from the same source. 5) Configure automatic restarts and health checks to limit downtime, and review issue #50927 for technical details.
What systems are affected by CVE-2026-103241?
This vulnerability affects the following AI/ML architecture patterns: model serving, LLM inference endpoints, RAG pipelines, agent frameworks.
What is the CVSS score for CVE-2026-103241?
CVE-2026-103241 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.71%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0029Denial of AI ServiceAML.T0040AI Model Inference API AccessAML.T0049Exploit Public-Facing ApplicationCompliance Controls Affected
What are the technical details?
Original Advisory
A flaw has been found in vllm-project vLLM up to 0.26.0. This vulnerability affects unknown code of the file rust/src/parser/src/unified/gemma4.rs of the component Gemma4UnifiedParser. Executing a manipulation can lead to denial of service. The attack may be launched remotely. The exploit has been published and may be used. Upgrading to version 0.29.1rc0 is able to resolve this issue. This patch is called 3439bad37e68ba9755a46f4f6b44a4aeaf1f60a9. Upgrading the affected component is advised.
Exploitation Scenario
An attacker finds an exposed vLLM endpoint (OpenAI-compatible API) serving a Gemma4 model. Using the public PoC, they send a crafted request that triggers the faulty handling in Gemma4UnifiedParser. The parser fails to release resources properly or enters an erroneous state, and the inference process becomes unresponsive or crashes. By repeating the request, the attacker keeps the service down, interrupting every business application that depends on the model.
Weaknesses (CWE)
CWE-404 — Improper Resource Shutdown or Release: The product does not release or incorrectly releases a resource before it is made available for re-use.
- [Requirements] Use a language that does not allow this weakness to occur or provides constructs that make this weakness easier to avoid. For example, languages such as Java, Ruby, and Lisp perform automatic garbage collection that releases memory for objects that have been deallocated.
- [Implementation] It is good practice to be responsible for freeing all resources you allocate and to be consistent with how and where you free memory in a function. If you allocate memory that you intend to free upon completion of the function, you must be sure to free the memory at all exit points for that function including error conditions.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:HReferences
- gist.github.com/Yunzez/8e98d656aa667095b513161eb056d28e
- github.com/vllm-project/vllm/
- github.com/vllm-project/vllm/commit/3439bad37e68ba9755a46f4f6b44a4aeaf1f60a9
- github.com/vllm-project/vllm/issues/50927
- github.com/vllm-project/vllm/pull/54303
- github.com/vllm-project/vllm/releases/tag/v0.29.1rc0
- vuldb.com/cve/CVE-2026-103241
- vuldb.com/submit/956250
- vuldb.com/vuln/411965
- vuldb.com/vuln/411965/cti
Timeline
Related Vulnerabilities
CVE-2026-6173210.0Decepticon: ChatML token injection yields sandbox RCE
Same package: vllmCVE-2024-110419.8vllm: RCE via unsafe pickle deserialization in MessageQueue
Same package: vllmCVE-2026-259609.8vllm: SSRF allows internal network access
Same package: vllmCVE-2025-472779.8vLLM: RCE via exposed TCPStore in distributed inference
Same package: vllmCVE-2024-90539.8vllm: RCE via unsafe pickle deserialization in RPC server
Same package: vllm