CVE-2026-54236 — vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic router
vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, the fix for CVE-2026-22778, which introduced a sanitize_message helper that strips object-repr memory addresses from error messages before they reach the client, is incomplete: several response paths echo str(exc) directly to clients without calling sanitize_message. The unsanitized sites include the Anthropic API router in vllm/entrypoints/anthropic/api_router.py (the POST /v1/messages and POST /v1/messages/count_tokens handlers), the Server-Sent Events streaming converter in vllm/entrypoints/anthropic/serving.py, and the realtime speech-to-text WebSocket in vllm/entrypoints/speech_to_text/realtime/connection.py. These paths catch the exception inside the route coroutine and construct the JSONResponse themselves, bypassing the sanitizing global FastAPI exception handler, and WebSocket frames do not traverse that handler chain at all. Using the same primitive as the parent issue, an unauthenticated attacker can send malformed image bytes through the Anthropic Messages API image content parts so that PIL.Image.open raises an UnidentifiedImageError whose message contains the BytesIO object repr, leaking the heap memory address verbatim in the error.message field of the response body. This vulnerability is fixed in 0.23.1rc0.
Who exploits it and how urgent it is
A 0–100 score combining confirmed exploitation, ransomware and APT use, public exploits, EPSS and severity.
How the score was calculated
- Public exploit/PoC (nuclei)+4
- EPSS 0.9% (30-day exploitation probability)+9
- Technical severity (medium)+3
- CISA SSVC: exploitation "poc"+5
- Automatable exploitation (SSVC)+3
The score orders the patch queue; it does not replace your asset context (internet exposure, criticality, compensating controls).
Triage
Is it exploited, how likely is exploitation, what does it touch, and how severe do the sources call it.
CISA SSVC assessment CISA
Three decision points CISA publishes for the CVEs it assesses (SSVC 2.0.3). A stakeholder decision, not a severity score.
Remediation
Fixes and mitigations documented by the sources — linked to the original advisory.
- Documented patch — github.com (NIST National Vulnerability Database)
First 24 hours
Ordered from the record’s own fields — exposure first, because you cannot patch what you have not found. A rule-generated starting point.
- Identify exposed assets running the affected vendor/product/version combinations (Vllm Vllm, Vllm-project Vllm-project/vllm).
- The attack is network-based and needs no privileges: restrict exposure (ACLs, VPN, WAF) until patched.
- Apply the fixed version or the patch documented under Remediation.
- Search available logs for exploit probes matching the vulnerability class (info disclosure), errors and authentication anomalies.
Affected scope
Vendor, product and versions as the sources word them.
| Vendor | Product | Versions | Status | Source |
|---|---|---|---|---|
| Vllm | Vllm | < 0.23.1 | Vulnerable | NIST National Vulnerability Database |
| Vllm-project | Vllm-project/vllm | version < 0.23.1rc0 | Vulnerable | CVE Program (cve.org) e CISA ADP |
Attack characteristics
The CVSS vector, decoded. It describes the attack, not your exposure to it.
Every base score collected
Sources score independently and may disagree; each row says who scored it and under which version.
| Score | Version | Severity | Expl. | Impact | Scored by |
|---|---|---|---|---|---|
| 5.3 | CVSS 3.1 | Medium | 3.9 | 1.4 | security-advisories@github.com via NIST National Vulnerability Database |
| 5.3 | CVSS 3.1 | Medium | — | — | GitHub_M via CVE Program (cve.org) e CISA ADP |
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N (score shown at the top: security-advisories@github.com) Weakness MITRE
Weakness class according to MITRE’s CWE catalog.
Public exploits, PoCs and detection templates
Public code that demonstrates or tests the flaw raises the odds of mass exploitation — and Nuclei templates help find vulnerable assets in your environment.
Warning: public PoCs are sometimes disguised malware. Use them only for authorized testing in an isolated environment. This portal does not host exploit code — it only points to the sources.
EPSS over time FIRST
Oct 13, 2026 → Oct 14, 2026 · 2 reading(s) · max 0.93%
Detection
Read off the CVSS vector and the weakness class. Starting points, not tested rules.
- Prioritize edge telemetry for the network-reachable component (proxy, WAF, IDS/IPS).
- Look for anomalous data read/export volumes and access to sensitive records.
Timeline
What happened to this CVE, newest first.
- 2026
- Affected products added · vllm-project/vllm
- SSVC assessment updated
- Record published
- Fix documented
- CVSS changed · 5.3
- Source status changed · → Analyzed
- Record created in this database
- CVE published
Source activity
Repeated readings are counted rather than listed.
- 1× Official description updated · Oct 13, 2026 – Oct 13, 2026
References
3 on the record
Elsewhere on this site
Related CVEs (same products)
- Medium CVE-2026-109311 vLLM before 0.29.0 Heap Overflow DoS via IAMF Audio Transcription Upload 3d
- Medium CVE-2026-105922 vllm-project vLLM Penalty utils.py get_token_bin_counts_and_mask denial of service 9d
- Medium CVE-2026-105775 vllm-project vLLM Completions Request mamba_mixer2.py conv_ssm_forward out-of-bounds 9d
- Medium CVE-2026-105756 vLLM: Loose `cache_salt` validation lets a single request kill EngineCore on LMCache-MP deployments — uncaught downstream `ValueError` denial of service 9d
- Medium CVE-2026-105758 vLLM: Qwen2-VL / Qwen3-VL video samplers bound on request-controlled max_frames, which the num_frames ceiling does not reach 9d
- Medium CVE-2026-105755 vLLM: Flash late-interaction scoring caches query embeddings under a caller-controlled request id — cross-request integrity break and induced errors on `/score` and `/rerank` 9d
Not in any source we poll
Listed rather than left blank: an empty field and an unmeasured one look identical on screen, and only one is a reason to look elsewhere.
- No confirmed IOCs, IP addresses, domains, file hashes, or malware artifacts supplied.
- No organization-specific asset inventory, compensating-control status, or patch deployment evidence supplied.
- No exploit packet captures, log samples, or incident case IDs supplied.
Answered from this record
Text assembled only from the fields above — no additional claims.
What should defenders know first?
CVE-2026-54236 is "vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic router", rated Medium. It affects Vllm (Vllm); versions are in the scope table. It is not in CISA KEV, which does not rule out exploitation. CVSS 5.3 describes technical severity. EPSS 0.93% estimates the probability of exploitation in 30 days; neither replaces your asset context. First verify whether exposed or business-critical assets run the affected versions, apply vendor fixes or mitigations, and preserve logs for review.
Comments
Moderated technical discussion. Read the community guidelines.
No comments yet.
Sign in or create an account to comment and suggest corrections.