BlackTreeIndependent security intelligence
← Back to the CVE catalogue
Full vulnerability report · 2026
CVE-2026-61539High confidence

Xinference: Remote code execution via unsafe `eval()` in Llama3 tool-call parsing

xorbitsai · inference

10.0CriticalCVSS 3.1
Recommended action
Within 72 hours

Critical technical impact with a remotely reachable, unauthenticated path; no CISA KEV confirmation is currently recorded.

Patch available
Optional official sources

National CERT insights
?CERT means Computer Emergency Response Team; CSIRT is the closely related term Computer Security Incident Response Team.

Select the national-authority views to include. The exact source language is shown on each matched advisory. Your choice is remembered on this device and encoded in the shareable URL.

Official European source

ENISA European Vulnerability Database

Official EUVD identifiers, advisory evidence and known-exploited context. Missing fields are not treated as evidence of low risk.

1 current
ENISA EUVD identifier

EUVD-2026-64146

No EUVD known-exploited evidence

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.

EUVD state
Present in the current official mapping
Known exploitation
Not present in the current ENISA EUVD known-exploited dataset. This is not proof of no exploitation.
ENISA score
10.0 · CVSS 3.1
Advisory evidence
No linked advisory details stored yet
Recommended actionWithin 72 hours

Critical technical impact with a remotely reachable, unauthenticated path; no CISA KEV confirmation is currently recorded.

Patch available
01

What, why and how

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.

What

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.

Why

The product receives input from an upstream component, but it does not neutralize or incorrectly neutralizes code syntax before using the input in a dynamic evaluation call (e.g. eval).

How

An attacker operating through a network path may attempt exploitation without authentication or user interaction. If successful, the issue may execute code or commands in the affected security context.

What

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.

Why

The product receives input from an upstream component, but it does not neutralize or incorrectly neutralizes code syntax before using the input in a dynamic evaluation call (e.g. eval).

How

An attacker operating through a network path may attempt exploitation without authentication or user interaction. If successful, the issue may execute code or commands in the affected security context.

02

Exploit reality and attack path

CVSS severity, EPSS forecast probability, public exploit material and CISA-confirmed exploitation are separate signals.

Observed exploitation
?Confirmed exploitation and public exploit material are separate signals. Attacks can occur without public proof-of-concept or exploit code.
No confirmed evidence

No CISA KEV match was present at the last successful refresh. This means no confirmation from that source, not proof of no exploitation.

Public PoC / exploit material
?Confirmed exploitation and public exploit material are separate signals. Attacks can occur without public proof-of-concept or exploit code.
None recorded

No exploit-tagged reference or CISA SSVC proof-of-concept state is currently recorded. Research may still exist outside the structured feeds.

Likely attack path
a network path → Improper Neutralization of Directives in Dynamically Evaluated Code ('Eval Injection') → execute code or commands in the affected security context
Attack surface
Network
Privileges required
None: unauthenticated exploitation is possible
User interaction
None
Attack complexity
Low: no specialised conditions are recorded
Security boundary
Changed: exploitation can affect a different security authority
Weakness
?CWE means Common Weakness Enumeration: a standard category for the underlying weakness.
CWE-95 ↗

CWE-95: Improper Neutralization of Directives in Dynamically Evaluated Code ('Eval Injection'). The product receives input from an upstream component, but it does not neutralize or incorrectly neutralizes code syntax before using the input in a dynamic evaluation call (e.g. eval).

CVSS vector
?CVSS means Common Vulnerability Scoring System. The vector records the metric values used to calculate technical severity.
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H

Common Vulnerability Scoring System 3.1: the compact vector below is decoded into plain language.

AVNetworkAttack vector: The vulnerable component can be reached over a network.ACLowAttack complexity: No specialised conditions are required beyond attacker-controlled input.PRNonePrivileges required: The attacker does not need an account or existing privileges.UINoneUser interaction: No action by another user is required.SChangedScope: The attack can affect a component governed by a different security authority.CHighConfidentiality impact: A successful attack can cause a major loss.IHighIntegrity impact: A successful attack can cause a major loss.AHighAvailability impact: A successful attack can cause a major loss.
Post-exploitation / living off the land
After compromise, an attacker may use built-in shells, scripting engines, scheduled tasks and native network utilities for discovery, persistence or movement. This is a plausible LoTL path, not evidence that it has occurred for every attack.
NetworkUnauthenticatedRemote code executionCWE-95
A

Official authority intelligence

Only matched European and national findings are included. Language selectors and unavailable sources are omitted.

ENISA EUVD · EUVD-2026-64146Official EUVD mapping

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.

Official EUVD record
03

Patch and workaround

Operational remediation based on structured source evidence.

Status
?Patch availability is based on structured fixed-version fields and authoritative update references. If no fix is verified, check the vendor advisory before making a change.
Patch available
Affected
< 2.7.0
Fixed
2.7.0.
Action
Review the linked authoritative reference and apply the recorded fixed release appropriate to the affected product branch.
Workaround
No verified workaround is recorded. If business-safe, reduce exposure to the affected interface and allow only trusted sources until authoritative guidance is available.
04

Evidence and provenance

Published 21 Aug 2026 · Last source change 24 Aug 2026, 19:45 UTC · CWE-95 · Improper Neutralization of Directives in Dynamically Evaluated Code ('Eval Injection')

CVE recordCVE.org · 5.2
CVSS sourceCNA
EPSS source
?The date BlackTree first stored a score for this CVE from the daily FIRST EPSS feed.
FIRST · tracked since 2026-08-23
European sourceENISA EUVD · EUVD-2026-64146
Product sourceCNA
Remediation sourceCVE/CNA references
CWE sourceCNA
NVD statusNVD received

Core structured fields are present and their contributing authorities are shown above.

Material change intelligence

What changed after publication

View recent updates →
  1. ENISA EUVD mappingEUVD-2026-64146 was added to the official ENISA EUVD mapping for this CVE.
    Before
    not recorded
    After
    {"euvdId":"EUVD-2026-64146"}
    ENISA EUVD ↗
  2. Catalogue recordCVE added to the BlackTree catalogue.
    CNA ↗
Material fields only · duplicate refreshes suppressed · history retained for the configured operational retention period
CVE published
Fixed release or patch reference recorded
ENISA EUVD · EUVD-2026-64146↗CVE authority or vendor reference↗Official CVE record (CNA and ADP data)↗NVD record↗
Technical terms and abbreviations used in this report
CVE
Common Vulnerabilities and Exposures: the public identifier for one disclosed vulnerability.
CVSS
Common Vulnerability Scoring System: a technical severity framework; it is not patching priority by itself.
EPSS
Exploit Prediction Scoring System: FIRST's estimate of the probability that exploitation activity will be observed in the next 30 days; it is a forecast, not confirmation.
CWE
Common Weakness Enumeration: the standard category describing the underlying software or hardware weakness.
CNA
CVE Numbering Authority: an organisation authorised to assign and publish CVE records.
CISA ADP
Cybersecurity and Infrastructure Security Agency Authorized Data Publisher: structured enrichment added to a CVE record.
NVD
National Vulnerability Database: NIST's enrichment service for CVE records.
CERT / CSIRT
A computer security incident response team that publishes warnings or coordinates incident response.
PoC
Proof of concept: public material that demonstrates or helps reproduce exploitation.
CSAF
Common Security Advisory Framework: a machine-readable format for security advisories.
LoTL
Living off the land: abuse of legitimate tools or system functions during an attack.
Free version - for non-commercial use only.CVE-2026-61539 · cve.blacktree.nl