← Vulnerability feed

Vulnerability record · CVE-2026-76841 · published 24 August 2026

CVE-2026-76841: Code injection vulnerability

Xinference loads models with Hugging Face remote code execution unconditionally enabled, and before version 2.12.0 exposes no setting to disable it. Six loader call sites pass trust_remote_code=True as a literal or as an unconditional default: RerankModel._get_tokenizer in xinference/model/rerank/core.py, SentenceTransformerRerankModel.load in xinference/model/rerank/sentence_transformers/core.py, SentenceTransformerEmbeddingModel.load in xinference/model/embedding/sentence_transformers/core.py, FlagEmbeddingModel.load in xinference/model/embedding/flag/core.py, and two sites in xinference/model/llm/transformers/core.py where PytorchModel._sanitize_model_config and PytorchModel._get_components default the value to True. Because a caller with model launch access can register a model whose type is unknown and supply an arbitrary model path, the server reaches _auto_detect_type and then AutoTokenizer.from_pretrained, which imports and executes Python declared by the model directory's own tokenizer_config.json auto_map, running attacker-supplied code with the privileges of the worker process. Version 2.12.0 gates every site behind allow_trust_remote_code and the XINFERENCE_TRUST_REMOTE_CODE setting, permitting remote code only for bundled built-in models.

8.7 CVSS 4.0 High EPSS 1.0% · top 37.7% CWE-94 · Code injection Deferred
8.7CVSS 4.0 base score
1.0%EPSS exploitation probability, 30 days
NoNot in CISA KEV
0Affected product versions listed by NVD
5References
24 Sep 2026Last modified by NVD

Description

Xinference loads models with Hugging Face remote code execution unconditionally enabled, and before version 2.12.0 exposes no setting to disable it. Six loader call sites pass trust_remote_code=True as a literal or as an unconditional default: RerankModel._get_tokenizer in xinference/model/rerank/core.py, SentenceTransformerRerankModel.load in xinference/model/rerank/sentence_transformers/core.py, SentenceTransformerEmbeddingModel.load in xinference/model/embedding/sentence_transformers/core.py, FlagEmbeddingModel.load in xinference/model/embedding/flag/core.py, and two sites in xinference/model/llm/transformers/core.py where PytorchModel._sanitize_model_config and PytorchModel._get_components default the value to True. Because a caller with model launch access can register a model whose type is unknown and supply an arbitrary model path, the server reaches _auto_detect_type and then AutoTokenizer.from_pretrained, which imports and executes Python declared by the model directory's own tokenizer_config.json auto_map, running attacker-supplied code with the privileges of the worker process. Version 2.12.0 gates every site behind allow_trust_remote_code and the XINFERENCE_TRUST_REMOTE_CODE setting, permitting remote code only for bundled built-in models.

CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/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:

References

LinkTags
https://github.com/xorbitsai/inference
https://github.com/xorbitsai/inference/blob/v2.11.0/xinference/model/rerank/core.py
https://github.com/xorbitsai/inference/issues/5023
https://github.com/xorbitsai/inference/pull/5027
https://www.vulncheck.com/advisories/xinference-through-remote-code-execution-via-hardcoded-trust-remote-code-in-model-l

Track CVE-2026-76841 inside VULONE

Watch it alongside the ransomware crews, C2 infrastructure and forum chatter that reference it, query it through the API and pull it into your SIEM over TAXII.

Start free Open in platform

Related vulnerabilities

Same products first, then exploited flaws of the same weakness class.

8.8CVE-2026-65660Microsoft sharepoint server code injection vulnerabilityImproper control of generation of code ('code injection') in Microsoft Office SharePoint allows an authorized attacker to execute code over a network.KEVEPSS 2.1%9.8CVE-2026-60004Gitea diffpatch API code injection enables remote code executionGitea before 1.27.1 allows remote code execution through the diffpatch API by way of Git hook installation. The flaw is a code injection issue (CWE-9…KEVEPSS 24%analysed9.5CVE-2026-72530TrueConf Server sandbox breakout via crafted script code injectionTrueConf Server versions 5.3.X through 5.3.9, 5.4.X through 5.4.9, 5.5.X through 5.5.5 and earlier allow a remote unauthenticated attacker to break o…KEVEPSS 1.7%analysed9.4CVE-2025-62593Ray browser-based RCE via insufficient User-Agent guardRay, an AI compute engine, contains a critical remote code execution flaw before version 2.52.0. Its defense against browser-based attacks relies on …KEVEPSS 62%analysed9.8CVE-2026-9198Langflow auto_login and code validation chain enables unauthenticated RCEIBM Langflow OSS 1.0.0 through 1.10.0 exposes /api/v1/auto_login, which mints SUPERUSER tokens to any network caller, and /api/v1/validate/code, whic…KEVEPSS 29%analysed7.2CVE-2026-15410SonicWall SMA1000 AMC code injection allows OS command executionThe SMA1000 Appliance Management Console (AMC) contains a post-authentication code injection flaw (CWE-94) that, under specific conditions, lets an a…KEVEPSS 12%analysed9.8CVE-2008-4250Microsoft Windows Server service RPC path canonicalization buffer overflowThe Server service in multiple Windows versions fails to properly handle path canonicalization, allowing a crafted RPC request to overflow a buffer a…KEVEPSS 99%analysed8.8CVE-2026-34197Apache ActiveMQ Jolokia JMX bridge code injection via broker config URIApache ActiveMQ Classic exposes the Jolokia JMX-HTTP bridge at /api/jolokia/ with a default policy that allows exec operations on all org.apache.acti…KEVEPSS 15%analysed

Source: NIST National Vulnerability Database (record CVE-2026-76841), CISA KEV, FIRST EPSS (scores of 2026-09-27). This page is refreshed as NVD updates the record.