CVE-2025-23275
Telemetry drift only. No threshold crossed and no authority ruling changed.
CVE record, CISA KEV and EPSS as of 2026-10-18.
CVEs like this one computed, by meaning
What to do now held facts and links
- Upgrade to cuda toolkit 13.0.0 or later upper bounds of the CVE record's affected ranges
- Record your decision, dated and frozen with today's evidence: patch by a date · mitigated by a compensating control · deferred · accepted the risk · not affected
- Check your version against the ranges this record publishes.
What it is stated by the CNA
NVIDIA CUDA Toolkit for all platforms contains a vulnerability in nvJPEG where a local authenticated user may cause a GPU out-of-bounds write by providing certain image dimensions. A successful exploit of this vulnerability may lead to denial of service and information disclosure.
- weakness
- CWE-787 Out-of-bounds Write
- affected
- NVIDIA CUDA Toolkit; NVIDIA nvJPEG
- published ranges
- nvidia cuda toolkit < 13.0.0
- first unaffected version, by branch
- cuda toolkit 13.0.0
- severity
- 4.2 CVSS:3.1/AV:L/AC:H/PR:L/UI:R/S:U/C:L/I:L/A:L
- assigned by
- nvidia
- published
- 2025-09-24
check us nvd.nist.gov cve.org nvidia.custhelp.com
Exploit code held, dated
- exploit code
- None held in the PoC-in-GitHub feed. A public repository changes this brief.
What moved held, dated
| date | EPSS | event |
|---|---|---|
| 25 Sep 2025 | 0.00012 | |
| 30 Sep 2025 | 0.00015 | |
| 7 Oct 2025 | 0.00014 | |
| 26 Oct 2025 | 0.00016 | |
| 1 Dec 2025 | 0.00019 | |
| 24 Jan 2026 | 0.00027 | |
| 20 Feb 2026 | 0.00013 | |
| 4 Mar 2026 | 0.00014 | |
| 20 Apr 2026 | 0.00016 | |
| 15 Jun 2026 | 0.00124 | scoring model v2026.06.15 |
| 20 Jun 2026 | 0.00127 | |
| 28 Aug 2026 | 0.00137 | |
| 14 Oct 2026 | 0.0014 |
13 observations held. A step marked with a scoring model moved because the measurement was replaced, not because the vulnerability changed.
Outside the record researched, cited
This section reads the record's references and adds cited findings from outside it. It needs an account. Everything above is held facts and is complete without it.
What would change this brief held, dated
- CISA KEV
- Not listed in the catalogue we hold. A listing changes this brief.
- exploit probability
- 0.00 on 2026-10-14 (FIRST's EPSS, a probability and not evidence of exploitation). Next band line: 10% (0.10). Crossing a line changes this brief.
- fix
- Held: cuda toolkit 13.0.0. A newer fix from the CNA or OSV changes this brief.
How to check this
Every value above is held with two dates: when it was true upstream (as_of) and when we recorded it (tx_from). Re-derive the exploit probability as it stood on any date:
SELECT value, source, source_version, as_of, tx_from
FROM volatile_valid_as_of('CVE-2025-23275', '2026-11-01T23:59:59.999999Z', now());