CVE-2026-7669 (GCVE-0-2026-7669)
Vulnerability from cvelistv5 – Published: 2026-05-02 22:00 – Updated: 2026-05-04 04:21
VLAI?
Title
sgl-project SGLang HuggingFace Transformer hf_transformers_utils.py get_tokenizer code injection
Summary
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 ?
Assigner
References
| URL | Tags | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
||||||||||||||
Impacted products
| Vendor | Product | Version | ||
|---|---|---|---|---|
| sgl-project | SGLang |
Affected:
0.5.0
Affected: 0.5.1 Affected: 0.5.2 Affected: 0.5.3 Affected: 0.5.4 Affected: 0.5.5 Affected: 0.5.6 Affected: 0.5.7 Affected: 0.5.8 Affected: 0.5.9 |
Credits
{
"containers": {
"cna": {
"affected": [
{
"modules": [
"HuggingFace Transformer Handler"
],
"product": "SGLang",
"vendor": "sgl-project",
"versions": [
{
"status": "affected",
"version": "0.5.0"
},
{
"status": "affected",
"version": "0.5.1"
},
{
"status": "affected",
"version": "0.5.2"
},
{
"status": "affected",
"version": "0.5.3"
},
{
"status": "affected",
"version": "0.5.4"
},
{
"status": "affected",
"version": "0.5.5"
},
{
"status": "affected",
"version": "0.5.6"
},
{
"status": "affected",
"version": "0.5.7"
},
{
"status": "affected",
"version": "0.5.8"
},
{
"status": "affected",
"version": "0.5.9"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "Nicholas Gould"
},
{
"lang": "en",
"type": "finder",
"value": "David Rochester"
},
{
"lang": "en",
"type": "reporter",
"value": "ngould (VulDB User)"
},
{
"lang": "en",
"type": "analyst",
"value": "ngould (VulDB User)"
},
{
"lang": "en",
"type": "coordinator",
"value": "VulDB CNA Team"
}
],
"descriptions": [
{
"lang": "en",
"value": "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\u0027s 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."
}
],
"metrics": [
{
"cvssV4_0": {
"baseScore": 6.3,
"baseSeverity": "MEDIUM",
"vectorString": "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/E:X",
"version": "4.0"
}
},
{
"cvssV3_1": {
"baseScore": 5.6,
"baseSeverity": "MEDIUM",
"vectorString": "CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:L/E:P/RL:X/RC:R",
"version": "3.1"
}
},
{
"cvssV3_0": {
"baseScore": 5.6,
"baseSeverity": "MEDIUM",
"vectorString": "CVSS:3.0/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:L/E:P/RL:X/RC:R",
"version": "3.0"
}
},
{
"cvssV2_0": {
"baseScore": 5.1,
"vectorString": "AV:N/AC:H/Au:N/C:P/I:P/A:P/E:POC/RL:ND/RC:UR",
"version": "2.0"
}
}
],
"problemTypes": [
{
"descriptions": [
{
"cweId": "CWE-94",
"description": "Code Injection",
"lang": "en",
"type": "CWE"
}
]
},
{
"descriptions": [
{
"cweId": "CWE-74",
"description": "Injection",
"lang": "en",
"type": "CWE"
}
]
}
],
"providerMetadata": {
"dateUpdated": "2026-05-04T04:21:53.695Z",
"orgId": "1af790b2-7ee1-4545-860a-a788eba489b5",
"shortName": "VulDB"
},
"references": [
{
"name": "VDB-360817 | sgl-project SGLang HuggingFace Transformer hf_transformers_utils.py get_tokenizer code injection",
"tags": [
"vdb-entry",
"technical-description",
"exploit"
],
"url": "https://vuldb.com/vuln/360817"
},
{
"name": "VDB-360817 | CTI Indicators (IOB, IOC, TTP, IOA)",
"tags": [
"signature",
"permissions-required"
],
"url": "https://vuldb.com/vuln/360817/cti"
},
{
"name": "Submit #799263 | sgl-project sglang \u003c=0.5.9 Protection Mechanism Failure",
"tags": [
"third-party-advisory"
],
"url": "https://vuldb.com/submit/799263"
},
{
"tags": [
"exploit"
],
"url": "https://github.com/gouldnicholas/CVE-2026-7669-PoC"
}
],
"timeline": [
{
"lang": "en",
"time": "2026-04-07T02:00:00.000Z",
"value": "Vulnerability found"
},
{
"lang": "en",
"time": "2026-04-07T02:00:00.000Z",
"value": "Vendor informed"
},
{
"lang": "en",
"time": "2026-05-02T00:00:00.000Z",
"value": "Advisory disclosed"
},
{
"lang": "en",
"time": "2026-05-02T02:00:00.000Z",
"value": "VulDB entry created"
},
{
"lang": "en",
"time": "2026-05-03T02:00:00.000Z",
"value": "Exploit disclosed"
},
{
"lang": "en",
"time": "2026-05-04T06:26:28.000Z",
"value": "VulDB entry last update"
}
],
"title": "sgl-project SGLang HuggingFace Transformer hf_transformers_utils.py get_tokenizer code injection"
}
},
"cveMetadata": {
"assignerOrgId": "1af790b2-7ee1-4545-860a-a788eba489b5",
"assignerShortName": "VulDB",
"cveId": "CVE-2026-7669",
"datePublished": "2026-05-02T22:00:19.859Z",
"dateReserved": "2026-05-02T08:00:13.701Z",
"dateUpdated": "2026-05-04T04:21:53.695Z",
"state": "PUBLISHED"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2",
"vulnerability-lookup:meta": {
"epss": {
"cve": "CVE-2026-7669",
"date": "2026-05-03",
"epss": "0.00035",
"percentile": "0.10097"
},
"nvd": "{\"cve\":{\"id\":\"CVE-2026-7669\",\"sourceIdentifier\":\"cna@vuldb.com\",\"published\":\"2026-05-02T22:16:24.080\",\"lastModified\":\"2026-05-04T06:16:01.440\",\"vulnStatus\":\"Received\",\"cveTags\":[],\"descriptions\":[{\"lang\":\"en\",\"value\":\"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\u0027s 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.\"}],\"metrics\":{\"cvssMetricV40\":[{\"source\":\"cna@vuldb.com\",\"type\":\"Secondary\",\"cvssData\":{\"version\":\"4.0\",\"vectorString\":\"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/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X\",\"baseScore\":6.3,\"baseSeverity\":\"MEDIUM\",\"attackVector\":\"NETWORK\",\"attackComplexity\":\"HIGH\",\"attackRequirements\":\"NONE\",\"privilegesRequired\":\"NONE\",\"userInteraction\":\"NONE\",\"vulnConfidentialityImpact\":\"LOW\",\"vulnIntegrityImpact\":\"LOW\",\"vulnAvailabilityImpact\":\"LOW\",\"subConfidentialityImpact\":\"NONE\",\"subIntegrityImpact\":\"NONE\",\"subAvailabilityImpact\":\"NONE\",\"exploitMaturity\":\"NOT_DEFINED\",\"confidentialityRequirement\":\"NOT_DEFINED\",\"integrityRequirement\":\"NOT_DEFINED\",\"availabilityRequirement\":\"NOT_DEFINED\",\"modifiedAttackVector\":\"NOT_DEFINED\",\"modifiedAttackComplexity\":\"NOT_DEFINED\",\"modifiedAttackRequirements\":\"NOT_DEFINED\",\"modifiedPrivilegesRequired\":\"NOT_DEFINED\",\"modifiedUserInteraction\":\"NOT_DEFINED\",\"modifiedVulnConfidentialityImpact\":\"NOT_DEFINED\",\"modifiedVulnIntegrityImpact\":\"NOT_DEFINED\",\"modifiedVulnAvailabilityImpact\":\"NOT_DEFINED\",\"modifiedSubConfidentialityImpact\":\"NOT_DEFINED\",\"modifiedSubIntegrityImpact\":\"NOT_DEFINED\",\"modifiedSubAvailabilityImpact\":\"NOT_DEFINED\",\"Safety\":\"NOT_DEFINED\",\"Automatable\":\"NOT_DEFINED\",\"Recovery\":\"NOT_DEFINED\",\"valueDensity\":\"NOT_DEFINED\",\"vulnerabilityResponseEffort\":\"NOT_DEFINED\",\"providerUrgency\":\"NOT_DEFINED\"}}],\"cvssMetricV31\":[{\"source\":\"cna@vuldb.com\",\"type\":\"Secondary\",\"cvssData\":{\"version\":\"3.1\",\"vectorString\":\"CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:L\",\"baseScore\":5.6,\"baseSeverity\":\"MEDIUM\",\"attackVector\":\"NETWORK\",\"attackComplexity\":\"HIGH\",\"privilegesRequired\":\"NONE\",\"userInteraction\":\"NONE\",\"scope\":\"UNCHANGED\",\"confidentialityImpact\":\"LOW\",\"integrityImpact\":\"LOW\",\"availabilityImpact\":\"LOW\"},\"exploitabilityScore\":2.2,\"impactScore\":3.4}],\"cvssMetricV2\":[{\"source\":\"cna@vuldb.com\",\"type\":\"Secondary\",\"cvssData\":{\"version\":\"2.0\",\"vectorString\":\"AV:N/AC:H/Au:N/C:P/I:P/A:P\",\"baseScore\":5.1,\"accessVector\":\"NETWORK\",\"accessComplexity\":\"HIGH\",\"authentication\":\"NONE\",\"confidentialityImpact\":\"PARTIAL\",\"integrityImpact\":\"PARTIAL\",\"availabilityImpact\":\"PARTIAL\"},\"baseSeverity\":\"MEDIUM\",\"exploitabilityScore\":4.9,\"impactScore\":6.4,\"acInsufInfo\":false,\"obtainAllPrivilege\":false,\"obtainUserPrivilege\":false,\"obtainOtherPrivilege\":false,\"userInteractionRequired\":false}]},\"weaknesses\":[{\"source\":\"cna@vuldb.com\",\"type\":\"Primary\",\"description\":[{\"lang\":\"en\",\"value\":\"CWE-74\"},{\"lang\":\"en\",\"value\":\"CWE-94\"}]}],\"references\":[{\"url\":\"https://github.com/gouldnicholas/CVE-2026-7669-PoC\",\"source\":\"cna@vuldb.com\"},{\"url\":\"https://vuldb.com/submit/799263\",\"source\":\"cna@vuldb.com\"},{\"url\":\"https://vuldb.com/vuln/360817\",\"source\":\"cna@vuldb.com\"},{\"url\":\"https://vuldb.com/vuln/360817/cti\",\"source\":\"cna@vuldb.com\"}]}}"
}
}
Loading…
Loading…
Experimental. This forecast is provided for visualization only and may change without notice. Do not use it for operational decisions.
Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
Loading…
Loading…