Finding and attribution
Structured-output exceptions can escape request-level error handling in three paths: the default auto backend, guidance with ngram_gpu speculative decoding, and the opt-in Rust frontend. A request triggering one of these paths can terminate the shared EngineCore and affect other tenants. The advisory groups the three paths as one request-boundary root cause and one finding.
Product: vLLM. AI system: GPT-5.5-Cyber. Team / organization: OpenAI / Trail of Bits.
CVE identifiers: CVE-2026-105757.
AI and human contribution (source account): The vLLM project advisory explicitly states that this vulnerability was discovered using GPT-5.5-Cyber as part of the Patch the Planet security initiative. This establishes AI-enabled discovery attribution; the source does not establish model autonomy or the private discovery date.
Human involvement is only described where the source account establishes it; unknown involvement does not mean none. Vendor confirmation of a flaw does not confirm every claim about AI contribution.
Discovery context: The advisory establishes public disclosure and the documented fix scope. It does not establish whether defenders knew of the issue or whether it was exploited before disclosure.
Remediation status: fix documented. The advisory describes affected versions below 0.30.0 and links proposed fix PR #51450; verify exact patched release before distribution-specific use.
A source-documented fix does not establish installed patches or current safety. Source checking does not independently reproduce the finding or establish observed exploitation.
Disclosed 23 Sep 2026. AI attribution dated 23 Sep 2026. Sources checked 16 Nov 2026; these are separate dates, not a live installation check.
Recorded CVSS: Medium · highest recorded score 6.5. Documented technical effects: denial of service.
Severity, consequences and exploitation evidence
Case grouping uses the highest recorded CVSS score, with its version-specific severity band. It is descriptive, not a combined or editorial risk index. All recorded assessments remain below; scores from different versions and sources may disagree. CVSS specifications define version-specific scoring and severity bands.
- CVE-2026-105757: CVSS 3.1 · 6.5 (Medium) ·
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H. vLLM GitHub Security Advisory · checked 16 Nov 2026.
CVE-2026-105757: denial of service.
Prerequisites / affected scope: Requires the corresponding structured-output backend/configuration and malformed or failure-inducing request: site 1 uses default auto backend; site 2 uses guidance plus ngram_gpu; site 3 uses the opt-in Rust frontend and an empty constraint. API-key middleware narrows site 3 to authenticated callers..
Exploitation evidence: The advisory describes reproducible conditions but does not report observed exploitation.. Impact source · checked 16 Nov 2026.
Documented fix: The advisory describes affected versions below 0.30.0 and links proposed fix PR #51450; verify exact patched release before distribution-specific use.