{"id":"CVE-2026-115911","aliases":["EUVD-2026-109996","GHSA-542m-9234-8gvf","PYSEC-2026-4344"],"status":"published","title":"vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread pool","description":"vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0.","severity":{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium","assessor":"cna","from":"cve"},"epss":{"day":"2026-11-12","score":0.00312,"percentile":0.24395,"model":"v2026.06.15"},"facts":{"in_kev":false,"known_exploited":false,"public_exploit":false,"fix_available":true},"cwes":["CWE-400","CWE-770","CWE-834"],"published":"2026-10-26T16:47:59Z","modified":"2026-11-01T02:30:57Z","fixes":[{"ecosystem":"PyPI","package":"vllm","fixed_in":["0.32.0"],"sources":["ghsa","osv"]}],"affected":[{"source":"ghsa","ecosystem":"PyPI","package":"vllm","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0.19.0"},{"fixed":"0.32.0"}]}]},{"source":"osv","ecosystem":"PyPI","package":"vllm","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0.19.0"},{"fixed":"0.32.0"}]}],"versions":["0.19.0","0.19.1","0.20.0","0.20.1","0.20.2","0.21.0","0.22.0","0.22.1","0.23.0","0.24.0","0.25.0","0.25.1","0.26.0","0.27.0","0.27.1","0.28.0","0.29.0","0.30.0","0.31.0"]},{"source":"osv","ecosystem":"PyPI","package":"vllm","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0.19.0"},{"fixed":"0.32.0"}]}],"versions":["0.19.0","0.19.1","0.20.0","0.20.1","0.20.2","0.21.0","0.22.0","0.22.1","0.23.0","0.24.0","0.25.0","0.25.1","0.26.0","0.27.0","0.27.1","0.28.0","0.29.0","0.30.0","0.31.0"]}],"products":[{"vendor":"vllm","product":"vllm"},{"vendor":"vllm-project","product":"vllm"}],"references":[{"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-542m-9234-8gvf","tags":["x_refsource_CONFIRM","Patch","Vendor Advisory","advisory","web","fix"]},{"url":"https://github.com/vllm-project/vllm/pull/50300","tags":["x_refsource_MISC","Patch","fix","web"]},{"url":"https://github.com/vllm-project/vllm/commit/cff08b461e086523cee8bba5bd3c66d302625134","tags":["x_refsource_MISC","Patch","fix","web"]},{"url":"https://github.com/vllm-project/vllm/releases/tag/v0.32.0","tags":["x_refsource_MISC","Release Notes","web","advisory"]},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2026-115911","tags":["advisory"]},{"url":"https://github.com/advisories/GHSA-542m-9234-8gvf"},{"url":"https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/115xxx/CVE-2026-115911.json","tags":["advisory"]},{"url":"https://github.com/vllm-project/vllm","tags":["package"]}],"sources":{"cve":[{"id":"CVE-2026-115911","title":"vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread pool","description":"vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0.","status":"published","published":"2026-10-26T16:47:59.377Z","modified":"2026-10-30T14:20:51.484Z","cvss":[{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium","assessor":"cna"}],"cwes":["CWE-400","CWE-770","CWE-834"],"updated_at":"2026-11-02T14:20:01.618694Z"}],"euvd":[{"id":"EUVD-2026-109996","description":"vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0.","status":"published","published":"2026-10-26T16:47:59Z","modified":"2026-10-30T14:20:51Z","cvss":[{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium"}],"updated_at":"2026-11-02T16:59:06.607412Z"}],"ghsa":[{"id":"GHSA-542m-9234-8gvf","title":"vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread pool","description":"### Summary\nvLLM renders chat templates by passing the resolved Jinja2 source to `tokenizer.apply_chat_template()`, which compiles and `.render()`s it in an `ImmutableSandboxedEnvironment`. The sandbox prevents code-execution escapes but imposes **no wall-clock timeout, no output-size cap, and no nested-loop or binop limit**. The render runs in a `ThreadPoolExecutor` whose `max_workers` defaults to **1**, and CPython worker threads cannot be interrupted. An authenticated API client (or a malicious model repo loaded with `trust_remote_code=False`) can therefore submit a single short template and permanently wedge **all** chat-style endpoints (`/v1/chat/completions`, `/v1/responses`, `/tokenize`, `/pooling`-chat, `/invocations`-chat) for every tenant until the process is restarted. A string-multiplication variant instead OOM-kills the API-server process. Because of a separately-reported (GHSA-jmjw-q4x3-hmc2) gate omission on `/v1/responses`, the remote authenticated attacker vector works in **default `vllm serve` configuration** with no operator opt-in.\n\n### Impact\nA single authenticated request of \u003c100 bytes permanently denies all chat-style inference (`/v1/chat/completions`, `/v1/responses`, `/tokenize`, `/pooling`-chat, `/invocations`-chat) to **every tenant** of the API-server replica until the operator restarts the process. The string-multiplication variant OOM-kills the process. In a multi-tenant deployment this is a cross-tenant denial-of-service from any one tenant. From the malicious-model angle, a model repo whose `tokenizer_config.json` carries the hostile template wedges every replica that loads it on the first legitimate chat request, with `trust_remote_code=False`.\n\n**Preconditions:** any non-harmony model (`model_type != gpt_oss`); `renderer_num_workers` at its default of 1; for the default-config authenticated API attacker vector, `--chat-template` must be unset (the default because most HF models ship their own).\n**Blast radius:** API-server process, per replica. Engine/worker processes unaffected by the CPU variant (event loop survives); affected by the memory variant (process OOM).\n**Persistence:** until process restart. Survives client disconnect.\n\nThis advisory was fixed in [MR50300](https://github.com/vllm-project/vllm/pull/50300).","status":"published","published":"2026-10-28T21:14:52Z","modified":"2026-10-28T21:14:54Z","cvss":[{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium"}],"cwes":["CWE-400","CWE-770","CWE-834"],"updated_at":"2026-11-02T14:41:39.926569Z"}],"nvd":[{"id":"CVE-2026-115911","description":"vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0.","status":"published","published":"2026-10-26T17:17:18.753Z","modified":"2026-10-30T15:16:45.86Z","cvss":[{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium","assessor":"secondary"}],"cwes":["CWE-400","CWE-770","CWE-834"],"updated_at":"2026-11-02T14:13:03.452847Z"}],"osv":[{"id":"CVE-2026-115911","title":"vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread pool","description":"vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0.","status":"published","published":"2026-10-26T16:47:59.377Z","modified":"2026-11-01T02:30:57.443898758Z","cvss":[{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium"}],"cwes":["CWE-400","CWE-770","CWE-834"],"updated_at":"2026-11-02T15:44:56.675758Z"},{"id":"GHSA-542m-9234-8gvf","title":"vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread pool","description":"### Summary\nvLLM renders chat templates by passing the resolved Jinja2 source to `tokenizer.apply_chat_template()`, which compiles and `.render()`s it in an `ImmutableSandboxedEnvironment`. The sandbox prevents code-execution escapes but imposes **no wall-clock timeout, no output-size cap, and no nested-loop or binop limit**. The render runs in a `ThreadPoolExecutor` whose `max_workers` defaults to **1**, and CPython worker threads cannot be interrupted. An authenticated API client (or a malicious model repo loaded with `trust_remote_code=False`) can therefore submit a single short template and permanently wedge **all** chat-style endpoints (`/v1/chat/completions`, `/v1/responses`, `/tokenize`, `/pooling`-chat, `/invocations`-chat) for every tenant until the process is restarted. A string-multiplication variant instead OOM-kills the API-server process. Because of a separately-reported (GHSA-jmjw-q4x3-hmc2) gate omission on `/v1/responses`, the remote authenticated attacker vector works in **default `vllm serve` configuration** with no operator opt-in.\n\n### Impact\nA single authenticated request of \u003c100 bytes permanently denies all chat-style inference (`/v1/chat/completions`, `/v1/responses`, `/tokenize`, `/pooling`-chat, `/invocations`-chat) to **every tenant** of the API-server replica until the operator restarts the process. The string-multiplication variant OOM-kills the process. In a multi-tenant deployment this is a cross-tenant denial-of-service from any one tenant. From the malicious-model angle, a model repo whose `tokenizer_config.json` carries the hostile template wedges every replica that loads it on the first legitimate chat request, with `trust_remote_code=False`.\n\n**Preconditions:** any non-harmony model (`model_type != gpt_oss`); `renderer_num_workers` at its default of 1; for the default-config authenticated API attacker vector, `--chat-template` must be unset (the default because most HF models ship their own).\n**Blast radius:** API-server process, per replica. Engine/worker processes unaffected by the CPU variant (event loop survives); affected by the memory variant (process OOM).\n**Persistence:** until process restart. Survives client disconnect.\n\nThis advisory was fixed in [MR50300](https://github.com/vllm-project/vllm/pull/50300).","status":"published","published":"2026-10-28T21:14:52Z","modified":"2026-10-29T11:10:38.03503501Z","cvss":[{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium"}],"cwes":["CWE-400","CWE-770","CWE-834"],"updated_at":"2026-11-02T15:46:37.864461Z"},{"id":"PYSEC-2026-4344","description":"vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0.","status":"published","published":"2026-10-26T17:17:18.753Z","modified":"2026-10-29T11:00:02.788141501Z","cvss":[{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium"}],"updated_at":"2026-11-02T15:52:57.671918Z"}]}}