{"id":"CVE-2026-22807","aliases":["EUVD-2026-3678","GHSA-2pc9-4j83-qjmr","PYSEC-2026-2010"],"status":"published","title":"vLLM affected by RCE via auto_map dynamic module loading during model initialization","description":"vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.","severity":{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high","assessor":"cna","from":"cve"},"epss":{"day":"2026-11-10","score":0.0098,"percentile":0.62522,"model":"v2026.06.15"},"facts":{"in_kev":false,"known_exploited":false,"public_exploit":true,"fix_available":true},"cwes":["CWE-94"],"published":"2026-01-21T16:12:54Z","modified":"2026-10-28T02:30:24Z","exploits":[{"source":"pocgithub","id":"otakuliu/CVE-2026-22807_Range","title":"otakuliu/CVE-2026-22807_Range","date":"2026-01-31","url":"https://github.com/otakuliu/CVE-2026-22807_Range"}],"fixes":[{"ecosystem":"PyPI","package":"vllm","fixed_in":["0.14.0"],"sources":["ghsa","osv"]}],"affected":[{"source":"ghsa","ecosystem":"PyPI","package":"vllm","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0.10.1"},{"fixed":"0.14.0"}]}]},{"source":"osv","ecosystem":"PyPI","package":"vllm","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0.10.1"},{"fixed":"0.14.0"}]}],"versions":["0.10.1","0.10.1.1","0.10.2","0.11.0","0.11.1","0.11.2","0.12.0","0.13.0"]},{"source":"osv","ecosystem":"PyPI","package":"vllm","ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0.10.1"},{"fixed":"0.14.0"}]}],"versions":["0.10.1","0.10.1.1","0.10.2","0.11.0","0.11.1","0.11.2","0.12.0","0.13.0"]}],"products":[{"vendor":"red hat","product":"red hat ai inference server"},{"vendor":"red hat","product":"red hat ai inference server 3.2"},{"vendor":"red hat","product":"red hat ai inference server 3.3"},{"vendor":"red hat","product":"red hat enterprise linux ai (rhel ai) 3"},{"vendor":"red hat","product":"red hat openshift ai (rhoai)"},{"vendor":"red hat","product":"red hat openshift ai 2.25"},{"vendor":"red hat","product":"red hat openshift ai 3.3"},{"vendor":"red hat","product":"red hat openshift ai 3.4"},{"vendor":"vllm","product":"vllm"},{"vendor":"vllm-project","product":"vllm"}],"references":[{"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-2pc9-4j83-qjmr","tags":["x_refsource_CONFIRM","Patch","Vendor Advisory","advisory","web"]},{"url":"https://github.com/vllm-project/vllm/pull/32194","tags":["x_refsource_MISC","Issue Tracking","Patch","fix","web"]},{"url":"https://github.com/vllm-project/vllm/commit/78d13ea9de4b1ce5e4d8a5af9738fea71fb024e5","tags":["x_refsource_MISC","Patch","fix","web"]},{"url":"https://github.com/vllm-project/vllm/releases/tag/v0.14.0","tags":["x_refsource_MISC","Product","Release Notes","web"]},{"url":"https://access.redhat.com/security/cve/CVE-2026-22807","tags":["vdb-entry","x_refsource_REDHAT","advisory","web"]},{"url":"https://bugzilla.redhat.com/show_bug.cgi?id=2431865","tags":["issue-tracking","x_refsource_REDHAT","report","web"]},{"url":"https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-22807.json","tags":["x_sadp-csaf-vex","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:3461","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:3462","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:30089","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:30088","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:30087","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:10184","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:42644","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:3782","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:3713","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://access.redhat.com/errata/RHSA-2026:5119","tags":["vendor-advisory","x_refsource_REDHAT","advisory","web"]},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2026-22807","tags":["advisory"]},{"url":"https://pypi.org/project/vllm","tags":["web","package"]},{"url":"https://github.com/vllm-project/vllm","tags":["package"]},{"url":"https://github.com/pypa/advisory-database/tree/main/vulns/vllm/PYSEC-2026-2010.yaml","tags":["web"]},{"url":"https://github.com/advisories/GHSA-2pc9-4j83-qjmr","tags":["advisory"]},{"url":"https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/22xxx/CVE-2026-22807.json","tags":["advisory"]}],"sources":{"cve":[{"id":"CVE-2026-22807","title":"vLLM affected by RCE via auto_map dynamic module loading during model initialization","description":"vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.","status":"published","published":"2026-01-21T21:13:11.894Z","modified":"2026-07-21T12:05:15.59Z","cvss":[{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high","assessor":"cna"},{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high","assessor":"adp"}],"cwes":["CWE-94","CWE-94"],"updated_at":"2026-11-02T13:39:05.585774Z"}],"euvd":[{"id":"EUVD-2026-3678","description":"vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.","status":"published","published":"2026-01-21T21:13:11Z","modified":"2026-07-21T12:05:15Z","cvss":[{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high"}],"updated_at":"2026-11-02T18:19:47.959089Z"}],"ghsa":[{"id":"GHSA-2pc9-4j83-qjmr","title":"vLLM affected by RCE via auto_map dynamic module loading during model initialization","description":"# Summary\n\nvLLM loads Hugging Face `auto_map` dynamic modules during model resolution **without gating on `trust_remote_code`**, allowing attacker-controlled Python code in a model repo/path to execute at server startup.\n\n---\n\n# Impact\n\nAn attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve **arbitrary code execution** on the vLLM host during model load.  \nThis happens **before any request handling** and does **not require API access**.\n\n---\n\n# Affected Versions\n\nAll versions where `vllm/model_executor/models/registry.py` resolves `auto_map` entries with `try_get_class_from_dynamic_module` **without checking `trust_remote_code`** (at least current `main`).\n\n---\n\n# Details\n\nDuring model resolution, vLLM unconditionally iterates `auto_map` entries from the model config and calls `try_get_class_from_dynamic_module`, which delegates to Transformers’ `get_class_from_dynamic_module` and **executes the module code**.\n\nThis occurs even when `trust_remote_code` is `false`, allowing a malicious model repo to embed code in a referenced module and have it executed during initialization.\n\n### Relevant code\n\n- `vllm/model_executor/models/registry.py:856` — auto_map resolution  \n- `vllm/transformers_utils/dynamic_module.py:13` — delegates to `get_class_from_dynamic_module`, which executes code\n\n---\n\n# Fixes\n\n* https://github.com/vllm-project/vllm/pull/32194\n\n# Credits\n\nReported by **bugbunny.ai**","status":"published","published":"2026-01-21T16:12:54Z","modified":"2026-07-21T12:32:15Z","cvss":[{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high"}],"cwes":["CWE-94"],"updated_at":"2026-11-02T14:40:49.410128Z"}],"nvd":[{"id":"CVE-2026-22807","description":"vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.","status":"published","published":"2026-01-21T22:15:49.077Z","modified":"2026-07-21T12:17:29.847Z","cvss":[{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high","assessor":"secondary"},{"version":"3.1","score":9.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H","level":"critical","assessor":"primary"},{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high","assessor":"secondary"}],"cwes":["CWE-94"],"updated_at":"2026-11-02T14:12:04.423757Z"}],"osv":[{"id":"CVE-2026-22807","title":"vLLM affected by RCE via auto_map dynamic module loading during model initialization","description":"vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.","status":"published","published":"2026-01-21T21:13:11.894Z","modified":"2026-10-28T02:30:24.973932166Z","cvss":[{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high"}],"cwes":["CWE-94"],"updated_at":"2026-11-02T15:44:58.749922Z"},{"id":"GHSA-2pc9-4j83-qjmr","title":"vLLM affected by RCE via auto_map dynamic module loading during model initialization","description":"# Summary\n\nvLLM loads Hugging Face `auto_map` dynamic modules during model resolution **without gating on `trust_remote_code`**, allowing attacker-controlled Python code in a model repo/path to execute at server startup.\n\n---\n\n# Impact\n\nAn attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve **arbitrary code execution** on the vLLM host during model load.  \nThis happens **before any request handling** and does **not require API access**.\n\n---\n\n# Affected Versions\n\nAll versions where `vllm/model_executor/models/registry.py` resolves `auto_map` entries with `try_get_class_from_dynamic_module` **without checking `trust_remote_code`** (at least current `main`).\n\n---\n\n# Details\n\nDuring model resolution, vLLM unconditionally iterates `auto_map` entries from the model config and calls `try_get_class_from_dynamic_module`, which delegates to Transformers’ `get_class_from_dynamic_module` and **executes the module code**.\n\nThis occurs even when `trust_remote_code` is `false`, allowing a malicious model repo to embed code in a referenced module and have it executed during initialization.\n\n### Relevant code\n\n- `vllm/model_executor/models/registry.py:856` — auto_map resolution  \n- `vllm/transformers_utils/dynamic_module.py:13` — delegates to `get_class_from_dynamic_module`, which executes code\n\n---\n\n# Fixes\n\n* https://github.com/vllm-project/vllm/pull/32194\n\n# Credits\n\nReported by **bugbunny.ai**","status":"published","published":"2026-01-21T16:12:54Z","modified":"2026-09-10T03:49:54.420040254Z","cvss":[{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high"}],"cwes":["CWE-94"],"updated_at":"2026-11-02T15:46:34.700741Z"},{"id":"PYSEC-2026-2010","title":"vLLM affected by RCE via auto_map dynamic module loading during model initialization","description":"# Summary\n\nvLLM loads Hugging Face `auto_map` dynamic modules during model resolution **without gating on `trust_remote_code`**, allowing attacker-controlled Python code in a model repo/path to execute at server startup.\n\n---\n\n# Impact\n\nAn attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve **arbitrary code execution** on the vLLM host during model load.  \nThis happens **before any request handling** and does **not require API access**.\n\n---\n\n# Affected Versions\n\nAll versions where `vllm/model_executor/models/registry.py` resolves `auto_map` entries with `try_get_class_from_dynamic_module` **without checking `trust_remote_code`** (at least current `main`).\n\n---\n\n# Details\n\nDuring model resolution, vLLM unconditionally iterates `auto_map` entries from the model config and calls `try_get_class_from_dynamic_module`, which delegates to Transformers’ `get_class_from_dynamic_module` and **executes the module code**.\n\nThis occurs even when `trust_remote_code` is `false`, allowing a malicious model repo to embed code in a referenced module and have it executed during initialization.\n\n### Relevant code\n\n- `vllm/model_executor/models/registry.py:856` — auto_map resolution  \n- `vllm/transformers_utils/dynamic_module.py:13` — delegates to `get_class_from_dynamic_module`, which executes code\n\n---\n\n# Fixes\n\n* https://github.com/vllm-project/vllm/pull/32194\n\n# Credits\n\nReported by **bugbunny.ai**","status":"published","published":"2026-07-07T16:03:19.426478Z","modified":"2026-07-07T17:47:19.616561721Z","cvss":[{"version":"3.1","score":8.8,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H","level":"high"}],"updated_at":"2026-11-02T15:52:55.537808Z"}]}}