CVE-2026-7669

CVE-2026-7669 is a medium-severity vulnerability with a CVSS 3.x base score of 5.6. It is not currently listed as actively exploited by CISA, and its EPSS exploit-prediction score is low. The underlying weakness is classified as CWE-74.

Key facts

Description

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.

Frequently asked questions

What is CVE-2026-7669?
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.
How severe is CVE-2026-7669?
CVE-2026-7669 has a CVSS 3.x base score of 5.6, rated medium severity. It is exploitable over network with high attack complexity, requires no privileges and no user interaction. Impact on confidentiality is low, integrity low, and availability low.
Is CVE-2026-7669 being actively exploited?
It is not currently listed in CISA's KEV catalog. Its EPSS exploit-prediction score is 0% (29th percentile), an estimate of the probability of exploitation in the next 30 days.
How do I fix CVE-2026-7669?
Review the linked vendor and NVD advisories for patched versions and mitigations, then upgrade or apply the recommended workaround.
Does CVE-2026-7669 have an EU (EUVD) identifier?
Yes. CVE-2026-7669 is tracked in the ENISA EU Vulnerability Database (EUVD) as EUVD-2026-26802.
When was CVE-2026-7669 published?
CVE-2026-7669 was published on 2026-05-02 and last updated on 2026-06-17.

References

Other CWE-74 (Improper Neutralization of Special Elements (Injection)) vulnerabilities

Browse all CWE-74 (Improper Neutralization of Special Elements (Injection)) vulnerabilities →