CSIRTS // UNIFIED SECURITY ADVISORY FEEDSYS ● ONLINE · POWERED BY INTELFUSIONS.COM

CVE-2025-46153: PyTorch before 3.7.0 has a bernoulli_p decompose function in decompositions.py even though it lacks full consistency with the eager CPU implementation, negatively affecting nn.Dropout1d, nn.Dropout2d, and nn.Dropout3d for fallback_random=True.

mediumCVSS 5.3CVE-2025-46153

CSIRTS triage

vendor: PyTorchproduct: PyTorchOtheraffected: before 3.7.0
What
The bernoulli_p decompose function in decompositions.py lacks consistency with eager CPU implementation, affecting nn.Dropout layers with fallback_random=True.
Who is affected
PyTorch users running versions before 3.7.0 using Dropout layers with fallback_random=True.
Urgency
Medium severity (CVSS 5.3); incorrect numerical behavior in dropout layers, not currently exploited.
Action
Upgrade PyTorch to version 3.7.0 or later.

AI-assisted analysis generated from the source advisory — verify against the original.

⚡ Watch PyTorch

Get an email when a new PyTorch advisory drops — max one per day, one-click unsubscribe.

Details

Source
Microsoft Security Response Center (INTL · vendor-psirt · site)
Severity
medium — CVSS 5.3
Published
2026-08-06
Exploitation
Not in CISA KEV at last sync

Original advisory: https://msrc.microsoft.com/update-guide/vulnerability/CVE-2025-46153

Exploitation outlook

EPSS (FIRST.org) estimates each CVE’s probability of exploitation in the next 30 days — here is the CSIRTS.com read on those numbers.

Referenced CVEs

CVECSIRTS overviewExternal
CVE-2025-46153coverage & exploitation statusNVD · CVE.org

More from Microsoft Security Response Center