{"id":"CVE-2026-84286","aliases":["EUVD-2026-97575","GHSA-mw4q-pwhm-35f7"],"status":"published","title":"OOB Read/Write via Negative Array Index in exllamav3_ext Kernel Table","description":"ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a CUDA kernel without first verifying that the index is within bounds. A malicious model can force the code to read and invoke an out‑of‑range function pointer, resulting in a reproducible crash. Under specific conditions involving certain host memory layouts, it may be possible for an attacker to exploit this vulnerability to achieve controlled indirect kernel dispatch during model load.","severity":{"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":"adp","from":"cve"},"epss":{"day":"2026-11-09","score":0.00589,"percentile":0.48447,"model":"v2026.06.15"},"facts":{"in_kev":false,"known_exploited":false,"public_exploit":false,"fix_available":false},"cwes":["CWE-125","CWE-20","CWE-787"],"published":"2026-10-12T17:01:16Z","modified":"2026-10-19T12:32:19Z","products":[{"vendor":"turboderp-org","product":"exllamav3"}],"references":[{"url":"https://kb.cert.org/vuls/id/369611"},{"url":"https://github.com/turboderp-org/exllamav3/pull/310"},{"url":"https://www.kb.cert.org/vuls/id/369611"},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2026-84286"},{"url":"https://github.com/advisories/GHSA-mw4q-pwhm-35f7"}],"sources":{"cve":[{"id":"CVE-2026-84286","title":"OOB Read/Write via Negative Array Index in exllamav3_ext Kernel Table","description":"ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a CUDA kernel without first verifying that the index is within bounds. A malicious model can force the code to read and invoke an out‑of‑range function pointer, resulting in a reproducible crash. Under specific conditions involving certain host memory layouts, it may be possible for an attacker to exploit this vulnerability to achieve controlled indirect kernel dispatch during model load.","status":"published","published":"2026-10-12T17:01:16.032Z","modified":"2026-10-19T11:04:47.147Z","cvss":[{"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":"adp"}],"cwes":["CWE-125","CWE-787","CWE-20"],"updated_at":"2026-11-02T14:20:08.009733Z"}],"euvd":[{"id":"EUVD-2026-97575","description":"ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a CUDA kernel without first verifying that the index is within bounds. A malicious model can force the code to read and invoke an out‑of‑range function pointer, resulting in a reproducible crash. Under specific conditions involving certain host memory layouts, it may be possible for an attacker to exploit this vulnerability to achieve controlled indirect kernel dispatch during model load.","status":"published","published":"2026-10-12T17:01:16Z","modified":"2026-10-19T11:04:47Z","cvss":[{"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"}],"updated_at":"2026-11-02T16:59:51.629031Z"}],"ghsa":[{"id":"GHSA-mw4q-pwhm-35f7","title":"ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a CUDA kernel without...","description":"ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a CUDA kernel without first verifying that the index is within bounds. A malicious model can force the code to read and invoke an out‑of‑range function pointer, resulting in a reproducible crash. Under specific conditions involving certain host memory layouts, it may be possible for an attacker to exploit this vulnerability to achieve controlled indirect kernel dispatch during model load.","status":"published","published":"2026-10-12T18:31:55Z","modified":"2026-10-19T12:32:19Z","cvss":[{"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"}],"cwes":["CWE-20"],"updated_at":"2026-11-02T15:13:56.45986Z"}],"nvd":[{"id":"CVE-2026-84286","description":"ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a CUDA kernel without first verifying that the index is within bounds. A malicious model can force the code to read and invoke an out‑of‑range function pointer, resulting in a reproducible crash. Under specific conditions involving certain host memory layouts, it may be possible for an attacker to exploit this vulnerability to achieve controlled indirect kernel dispatch during model load.","status":"published","published":"2026-10-12T17:17:35.873Z","modified":"2026-10-19T12:17:10.86Z","cvss":[{"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":"secondary"}],"cwes":["CWE-20","CWE-125","CWE-787"],"updated_at":"2026-11-02T14:12:56.310333Z"}]}}