SGLang contains a RCE vulnerability when attempting to load model weights from a HuggingFace repository, specifically within the… (CVE-2026-15976)
SGLang contains a RCE vulnerability when attempting to load model weights from a HuggingFace repository, specifically within the /update_weights_from_disk, where torch.load(..., weights_only=False) fallback enables pickle deserialization of .bin files.
AI Analysis
Technical Summary
The vulnerability in SGLang arises from the use of torch.load with weights_only=false during the loading of model weights from a HuggingFace repository. This fallback enables pickle deserialization of .bin files, which can be exploited to achieve remote code execution. The flaw is specifically located in the /update_weights_from_disk endpoint or function. No patch or remediation details are currently provided, and there is no indication that this is a cloud service vulnerability.
Potential Impact
An attacker who can supply or influence the model weights loaded by SGLang can execute arbitrary code on the affected system due to unsafe deserialization of .bin files. This can lead to full system compromise depending on the privileges of the running process. There are no known exploits in the wild at this time.
Mitigation Recommendations
Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until a fix is available, avoid loading model weights from untrusted sources or repositories. Restrict access to the /update_weights_from_disk functionality to trusted users only.
SGLang contains a RCE vulnerability when attempting to load model weights from a HuggingFace repository, specifically within the… (CVE-2026-15976)
Description
SGLang contains a RCE vulnerability when attempting to load model weights from a HuggingFace repository, specifically within the /update_weights_from_disk, where torch.load(..., weights_only=False) fallback enables pickle deserialization of .bin files.
CVSS v3.1
Score 9.8critical
Weaknesses
AI-Powered Analysis
Machine-generated threat intelligence
Technical Analysis
The vulnerability in SGLang arises from the use of torch.load with weights_only=false during the loading of model weights from a HuggingFace repository. This fallback enables pickle deserialization of .bin files, which can be exploited to achieve remote code execution. The flaw is specifically located in the /update_weights_from_disk endpoint or function. No patch or remediation details are currently provided, and there is no indication that this is a cloud service vulnerability.
Potential Impact
An attacker who can supply or influence the model weights loaded by SGLang can execute arbitrary code on the affected system due to unsafe deserialization of .bin files. This can lead to full system compromise depending on the privileges of the running process. There are no known exploits in the wild at this time.
Mitigation Recommendations
Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until a fix is available, avoid loading model weights from untrusted sources or repositories. Restrict access to the /update_weights_from_disk functionality to trusted users only.
Technical Details
- Gcve Source
- db.gcve.eu
- Osv Id
- GHSA-r344-357p-w9pp
- Osv Schema Version
- 1.4.0
- Aliases
- ["CVE-2026-15976"]
Threat ID: 6a6bdd8a9c2644c7f8da71e8
Added to database: 07/30/2026, 23:26:02 UTC
Last enriched: 07/30/2026, 23:32:26 UTC
Last updated: 09/13/2026, 22:01:32 UTC
Views: 49
Community Reviews
0 reviewsCrowdsource mitigation strategies, share intel context, and vote on the most helpful responses. Sign in to add your voice and help keep defenders ahead.
Want to contribute mitigation steps or threat intel context? Sign in or create an account to join the community discussion.
Actions
Updates to AI analysis require Pro Console access. Upgrade inside Console → Billing.
Need more coverage?
Upgrade to Pro Console for AI refresh and higher limits.
For incident response and remediation, OffSeq services can help resolve threats faster.
Latest Threats
Check if your credentials are on the dark web
Instant breach scanning across billions of leaked records. Free tier available.