cve.bz

CVE-2026-7669 sgl-project sglang: Injection

6.3 mediumEPSS 0.4%

Medium (6.3) in sgl-project sglang. EPSS 0.4% estimated probability of exploitation activity in the next 30 days.

On this page
A vulnerability was detected in sgl-project SGLang up to 0.5.9. Impacted is the function get_tokenizer of the file python/sglang/srt/utils/hf_transformers_utils.py of the component HuggingFace Transformer Handler. The manipulation of the argument trust_remote_code with the input False as part of Boolean results in code injection. The attack can be executed remotely. A high complexity level is associated with this attack. The exploitability is considered difficult. In get_tokenizer(), when the caller passes trust_remote_code=False and HuggingFace transformers v5 returns a TokenizersBackend instance (the generic fallback for tokenizer classes not in the registry), SGLang silently re-invokes AutoTokenizer.from_pretrained with trust_remote_code=True, overriding the caller's explicit security setting. A model repository containing a malicious tokenizer.py referenced via auto_map in tokenizer_config.json will execute arbitrary Python in the SGLang process during this second call. No log line or warning is emitted. The override affects all current SGLang versions because transformers==5.3.0 is pinned in pyproject.toml. Both tokenizer_mode="auto" and tokenizer_mode="slow" are affected. The exploit is now public and may be used. The vendor was contacted early about this disclosure but did not respond in any way.

Severity

CVSS 4.0

6.3 medium
CVSS:4.0/AV:N/AC:H/AT:N/PR:N/UI:N/VC:L/VI:L/VA:L/SC:N/SI:N/SA:N

CVSS 3.1

5.6 medium
CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:L
EPSS
0.4% (unchanged from 0.4% a week ago; 0.4% 30 days ago) estimated probability of exploitation activity in the next 30 days; 31st percentile. Scored 2026-11-17. A forecast for this CVE across the internet, not for your own systems.
CISA's assessment (SSVC)
Exploitation: PoC · Automatable: no · Technical impact: partial, assessed 2026-05-05. From CISA's Vulnrichment in the CVE List: whether it is exploited, whether exploiting it can be automated, and how much control it gives.
CWEs
CWE-74 Improper Neutralization of Special Elements in Output Used by a Downstream Component (Injection)
CWE-94 Improper Control of Generation of Code (Code Injection)
OWASP Top 10
A05:2025 Injection
Published
Modified
NVD analysis
Deferred: the NVD does not plan to analyze it
Record from
the NVD, synced

Ask an AI about this CVE

Show the promptExplain CVE-2026-7669 for an engineer deciding whether to act. Use https://cve.bz/api/public/cves/CVE-2026-7669?via=ask-ai-copy and https://cve.bz/cve/CVE-2026-7669?via=ask-ai-copy as the source of truth: what is affected and which versions fix it, whether it is in CISA KEV, public exploits, and its EPSS (an internet-wide 30-day estimate of exploitation activity, not a forecast for my systems). If they give no fixed versions, use the CVE record https://cveawg.mitre.org/api/cve/CVE-2026-7669 or the vendor advisory, not third-party mirrors. Tell me how to check whether I am affected and how to verify a fix. Cite the sources and say where they are uncertain.

Opens a third-party assistant; its answer is its own, not cve.bz's. The prompt holds only this CVE's id and public links.

Affected

Versions

From the CNA's record in the CVE List.

ProductAffected
sgl-project SGLang0.5.0
sgl-project SGLang0.5.1
sgl-project SGLang0.5.2
sgl-project SGLang0.5.3
sgl-project SGLang0.5.4
sgl-project SGLang0.5.5
sgl-project SGLang0.5.6
sgl-project SGLang0.5.7
sgl-project SGLang0.5.8
sgl-project SGLang0.5.9

Packages (OSV)

AdvisoryEcosystemPackageAffectedFixed in
GHSA-6m5f-673f-5vh7PyPIsglang<=0.5.9no fix published
PYSEC-2026-3062PyPIsglang<=0.5.9no fix published

GitHub PoC

Repositories on GitHub named after this CVE or listed for it by the PoC-in-GitHub index. For those marked Harbor Check: matches CVE, Flawharbor read the code and found it implements this CVE's exploitation technique: read, not run, and its dependencies only checked for warning signs, so a pointer and no guarantee. The rest are unverified: some hold a working proof of concept, some nothing useful, and some malware made to look like one. Read the code before you run any of it. How Harbor Check reads them, and what it found.

  • gouldnicholas/CVE-2026-7669-PoC Harbor Check: matches CVE 1 star, Python, created
    PoC for CVE-2026-7669: SGLang silent trust_remote_code override -> RCE

Attack patterns (CAPEC)

  • CAPEC-35 Leverage Executable Code in Non-Executable Files Likelihood: High · Severity: Very High
  • CAPEC-77 Manipulating User-Controlled Variables Likelihood: High · Severity: Very High
  • CAPEC-242 Code Injection Likelihood: High · Severity: High

The attack patterns MITRE's CAPEC relates to this record's weaknesses, as MITRE names them. Credits. More on the page of CWE-74 (37).

References

Check your stackJSON