CVE-2026-86793: SGLang · Code Injection
Affects SGLangSource CVE Database V5
Public exploit available · CVSS 9.8 Critical : upgrade to 0.5.21 as a priority.
Data as of 16 Nov 2026, 17:34 UTC
Thin evidence
One scorer or none| CISA-ADP | CVSS 3.1 | 9.8 |
|---|
- CISA SSVC
- Exploitation
- Proof of concept is public or traded
- Automatable
- Attackers can reliably automate steps 1–4 of the kill chain
- Technical impact
- Total control over the software or total information disclosure
CISA SSVC assessment, 2026-09-14.
- Patch
- Available – fixed in 0.5.21
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:HExploitation evidence
t = 0 is the CVE’s publication on cve.org
Disclosed: evidence found.
CVE published 11 Sep 2026
Detectable: no evidence on record.
No public nuclei template indexed
Weaponized: evidence found.
CISA SSVC: proof of concept public
Exploited in the wildExploited: no evidence on record.
Not in CISA KEV and no source reports exploitation
Observed by OffSeq MirageSeen by Mirage: no evidence on record.
No attack traffic recorded by OffSeq Mirage sensors
Timeline
CVE ID reserved
by CERT/CC
, 3 days before disclosure
CVE published
on cve.org
, t = 0
CISA SSVC assessment
proof of concept public
, 3 days after disclosure
A missing step is not evidence of safety. Sources: cve.org (opens in a new tab), CISA SSVC, the Metasploit, Exploit-DB and nuclei-templates index as of 15 Nov 2026.
What to do
Upgrade SGLang to 0.5.21.
Affected software
| Product | Affected | Fixed in |
|---|---|---|
| SGLangSGLang | Affected
| Fixed in
|
AI-suggested
Upgrade SGLang to version 0.5.21 or later, where this vulnerability is fixed. Do not run vulnerable versions without authentication keys configured, as this enables exploitation. Since this is not a cloud service, remediation requires applying the vendor's fixed version.
Description
CVE-2026-86793 is a critical code injection vulnerability in SGLang versions before 0.5.21. It allows unauthenticated remote code execution via pickle deserialization through the /update_weights_from_tensor endpoint when no authentication keys are configured. The SafeUnpickler policy intended to prevent unsafe deserialization can be bypassed because builtins.import and builtins.getattr remain accessible, enabling malicious pickle REDUCE operations.