Advisory record · GHSA-jhr6-gm9c-rqjv

sentence-transformers local model loading bypasses trust_remote_code and executes custom Python

CVSS 9.3 CriticalCriticalEPSS 0.0091 · 60th percentile

Critical, because:

  • CVSS score 9.3: 9.0 or more.

How we decide what’s critical

Published by the GitHub Advisory Database at 31 Jul 2026, 21:31 UTC. agentprime has not analysed this yet, so whether it affects any app is not known.

PackageEcosystemAffectedFixed in
sentence-transformersPyPI<5.6.05.6.0
Identifiers
GHSA-jhr6-gm9c-rqjv · CVE-2026-68770 · PYSEC-2026-4164
Weakness (CWE)
CWE-94
Published
31 Jul 2026
Updated
1 Oct 2026

Machine-written summary

Summary written by Claude Haiku from the source advisory. Not reviewed by a person.

Affected: sentence-transformers versions before 5.6.0. Loading a local model directory with trust_remote_code=False still runs custom Python files from that directory, because a logic flaw in the import helper treats any existing local path as trusted. An attacker who can control or modify the contents of a model directory can get code executed in the process that loads the model. Upgrade to sentence-transformers 5.6.0 or later. The source gives no other workaround.

Written 1 Nov 2026.

Not analysed yet

We haven’t published our analysis of this Advisory yet. Once a person has reviewed it, this page will say what’s vulnerable, what has to be true for your app to be affected, and how to check.

Severity and exploitation

ScoreRatingCVSSVectorSource
9.3 Critical 4.0 CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N CNA (VulnCheck)
  • CISA KEV Not listed.
  • EPSS 0.0091, the 60th percentile, on 13 Nov 2026: FIRST’s estimate of the chance of exploitation in the next 30 days.

Timeline

  1. 31 Jul 2026, 20:56 UTC Confirmed by the CVE List
  2. 31 Oct 2026, 14:09 UTC First seen by agentprime
  3. 31 Oct 2026, 14:09 UTC Confirmed by OSV

Sources

Facts on this page come from the GitHub Advisory Database (CC-BY 4.0), OSV and the other sources listed, as published. We don’t reproduce full advisory descriptions, which can contain exploit code.