Prioritization
- NVD CVSS
- 10.0
- Published
- Aug 21, 2026 5:17 PM EDT
- Modified
- Aug 21, 2026 5:17 PM EDT
- CISA KEV
- No
← Threat Center
10.0
CRITICAL • CVSS 3.1
CVE-2026-61539
Xinference is an inference API for running open-source, speech, and multimodal models. In 2.5.0 and earlier, Xinference passes attacker-influenced Llama3 tool-call output to eval() in xinference/model/llm/tool_parsers/llama3_tool_parser.py and xinference/model/llm/utils.py. Requests to /v1/chat/completions with a tools field flow through xinference/api/restful_api.py, xinference/model/llm/transformers/core.py, handle_chat_result_non_streaming(), and _post_process_completion() before extract_tool_calls() or _eval_llama3_chat_arguments() evaluates the model-generated Python expression. An unauthenticated remote attacker can influence that output through a crafted prompt and execute commands in the Xinference server process context. This issue is fixed in version 2.7.0.
Red Hat context
No Red Hat-specific cached context is available for this record.
Recommended response
- Identify affected assets with Qualys, Tanium, software inventory, CMDB or configuration-management data.
- Confirm package/application applicability before change execution; CVSS alone does not prove that every host is vulnerable.
- Patch, upgrade, rehydrate or apply a vendor-approved mitigation through controlled change. GraceITS can automate eligible waves with Ansible/Tanium.
- Rescan after remediation and reconcile remaining exceptions, false positives, unsupported systems and compensating controls.