Four vLLM advisories published in one batch, including a bypass of an incomplete CVE-2025-62164 fix

GitHub Security Advisories·medium signalXBlueskyLinkedInCopy link

GitHub published four vLLM advisories on 2026-09-04, all medium severity, against the inference server most self-hosted agent stacks sit on. GHSA-pr7f-p5mw-fc87 shows the CVE-2025-62164 remediation can be bypassed using concurrent prompt parts; GHSA-48jh-3gj7-fg8v is a ReDoS through structured_outputs.regex in the lm-format-enforcer backend with no compile timeout, described as a missed sibling of GHSA-rwxx-mrjm-wc2m; GHSA-hwrm-c4cx-rf4j leaks internal paths and the username through unauthenticated validation error messages; and GHSA-8737-qx52-hjff has derender endpoints decoding caller-supplied token IDs with no output bounds. The regex one matters most for agent builders, since structured outputs are exactly how tool-calling gets constrained.