Read the docs

Predictable temp path in TorchServe enables local privilege escalation

Priority HIGH 69CWE-379Published · Updated

Key facts

Attack vector
Local
Privileges required
Low
User interaction
None
Attack complexity
Low
Impact type
—
Confidentiality · integrity · availability
High · High · High
Fix status
No fix listed

Attack profile from CVSS 3.1 · CNA. Fix status from CNA.

Summary

All versions of PyTorch TorchServe are affected. The extraction directory in /tmp is predictable, a flaw tracked as CWE-379. A local user with low privileges can swap the model handler in that directory and run code as the TorchServe service account.

Written with Claude Haiku 5.5 from the sources below · 2026-11-06

Description from the CNA

In all versions of PyTorch's TorchServe, a predictable extraction directory in /tmp lets a local user swap the model handler and run code as the TorchServe service account leading to local privilege escalation.

Why it matters now

Exploitation Proof of concept CISA SSVC · not in CISA KEV
EPSS 0.00145 3rd percentileScored 2026-11-07
Fix No fix listed No fixed version in the sources yet

SSVC decision points

CISA ADP · SSVC 2.0.3

Exploitation
Poc
Automatable
No
Technical impact
Total

The decision (Track, Attend, Act) also depends on your mission and well-being impact; see CISA's SSVC guide.

Affected products

VendorProductAffected versionsFixed inStatusSource
pytorchTorchServeall versions—No fix listedCNA
  • TorchServe Vendorpytorch Affected all versions Fixed in — No fix listed CNA

Remediation

The sources list no fixed version for PyTorch TorchServe.

Written with Claude Haiku 5.5 from the sources below · 2026-11-06

Timeline

  1. 2026-11-06 Threatist record created from CVE-2026-123974
  2. 2026-11-06 CISA SSVC decision points changed
  3. 2026-11-06 Priority changed: MEDIUM → HIGH
  4. 2026-11-06 Alias added: GHSA-g2fw-m3ph-hqcq

References

  • TI-2023-23379 TorchServe Server-Side Request Forgery CVSS 3.1CRITICAL 10.0
  • TI-2026-46956 PyTorch torchvision GIF Decoder Out-of-bounds Heap Read CVSS 4.0HIGH 7.1
  • TI-2024-06099 Unclaimed S3 Bucket Usage in pytorch/serve CVSS 3.0MEDIUM 6.3
  • TI-2025-23505 PyTorch: `torch.load` with `weights_only=True` leads to remote code execution CVSS 4.0CRITICAL 9.3
  • TI-2024-23271 TorchServe gRPC Port Exposure CVSS 3.1HIGH 8.2
  • TI-2024-23270 TorchServe bypass allowed_urls configuration CVSS 3.1CRITICAL 9.8
See all 8 →

Sources