Cyber Resilience

CVE-2026-4963

Huggingface Smolagents 1.25.0

Public PoC
Published
27 March 2026
Modified
30 April 2026
CVSS Score v4 2.1
Click a component to see what it means
Raw vectorCVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:L/VI:L/VA:L/SC:N/SI:N/SA:N/E:P/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
EPSS Score 0.0057 44th percentile
Risk Priority 15 floored blend · peak EPSS

Summary

CVE-2026-4963 is a low-severity Injection (CWE-74) vulnerability in Huggingface Smolagents. Its CVSS base score is 2.1 (Low).

Operationally, exploitation aligns with the MITRE ATT&CK technique Exploit Public-Facing Application (T1190); ranked at the 44th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog; a public proof-of-concept is referenced.

This vulnerability is AI-related — categorised as AI Agent Protocols and Integrations; in the LLM/Generative AI Risks risk domain.

The strongest mitigations our analysis identified map to SA-11 (Developer Testing and Evaluation) and SI-10 (Information Input Validation) — see the control section below for these in your framework.

Deeper analysis AI-assisted summary

Synthesised by an AI model from the NVD description and linked references — a reading aid, not an authoritative source.

CVE-2026-4963 is a code injection vulnerability in Hugging Face's smolagents library version 1.25.0.dev0, stemming from an incomplete fix for CVE-2025-9959. The issue affects the functions evaluate_augassign, evaluate_call, and evaluate_with within the file src/smolagents/local_python_executor.py. It enables manipulation that leads to arbitrary code execution, as classified under CWE-74 (Improper Neutralization of Special Elements in Output Used by a Downstream Component) and CWE-94 (Improper Control of Generation of Code).

The vulnerability can be exploited remotely over the network with low complexity and no required privileges, though it necessitates user interaction, such as clicking a malicious link or input. Successful exploitation grants low-impact confidentiality, integrity, and availability effects, with an overall CVSS v3.1 base score of 6.3 (AV:N/AC:L/PR:N/UI:R/S:U/C:L/I:L/A:L). Attackers can inject and execute code in the context of the affected Python executor.

Advisories from VulDB indicate that the vendor was contacted early regarding disclosure but provided no response, and no patches or mitigations are mentioned. Publicly available exploits, hosted on GitHub Gists, demonstrate the attack vectors.

Notable context includes the public availability of exploits, increasing the risk of real-world attacks against deployments using this development version of smolagents, which is part of Hugging Face's AI/ML agent ecosystem.

OWASP Top 10 for Web (2025)

EU & UK References

Vulnerability Data

A weakness has been identified in huggingface smolagents 1.25.0.dev0. This affects the function evaluate_augassign/evaluate_call/evaluate_with of the file src/smolagents/local_python_executor.py of the component Incomplete Fix CVE-2025-9959. This manipulation causes code injection. It is possible to initiate the attack remotely. The exploit has…

more

been made available to the public and could be used for attacks. The vendor was contacted early about this disclosure but did not respond in any way.

CWE(s)

AI Security AnalysisAI

AI Category
AI Agent Protocols and Integrations
Risk Domain
LLM/Generative AI Risks
OWASP Top 10 for LLMs 2025
None mapped
AI-specific weaknesses CR
  • CWE-1426 — Model-generated code reaches Python exec sink without validation (smolagents executor).
Mapped by Cyber Resilience · not in NVD. Poisoning and extraction cases are routed to MITRE ATLAS instead of a synthetic CWE.
Classification Reason
Matched keywords: huggingface

Related Threats

MITRE ATT&CK Enterprise Techniques

T1190 Exploit Public-Facing Application Initial Access
Adversaries may attempt to exploit a weakness in an Internet-facing host or system to initially access a network.
T1221 Template Injection Stealth
Adversaries may create or modify references in user document templates to conceal malicious code or force authentication attempts.
T1659 Content Injection Initial Access
Adversaries may gain access and continuously communicate with victims by injecting malicious content into systems through online network traffic.
T1674 Input Injection Execution
Adversaries may simulate keystrokes on a victim’s computer by various means to perform any type of action on behalf of the user, such as launching the command interpreter using keyboard shortcuts, typing an inline script to be executed,…
T1059 Command and Scripting Interpreter Execution
Adversaries may abuse command and script interpreters to execute commands, scripts, or binaries.
T1059.001 PowerShell Execution
Adversaries may abuse PowerShell commands and scripts for execution.
Derived from this CVE’s CWE(s) via the direct CWE→ATT&CK cross-walk.

CVEs Like This One

CVE-2025-5120Same product: Huggingface Smolagents
CVE-2025-14927Same vendor: Huggingface
CVE-2025-14928Same vendor: Huggingface
CVE-2025-14926Same vendor: Huggingface
CVE-2026-44827Same vendor: Huggingface
CVE-2026-4515Shared CWE-74, CWE-94
CVE-2026-5970Shared CWE-74, CWE-94
CVE-2026-1977Shared CWE-74, CWE-94
CVE-2026-6110Shared CWE-74, CWE-94
CVE-2026-16204Shared CWE-74, CWE-94

Affected Assets

huggingface
smolagents
1.25.0

Mitigating Controls

Control response

Prevent
Stop it (NIST 800-53)

Detect
Catch it (NIST detect / respond)

Harden
Shrink the surface (DISA STIG)

Validate
Prove the fix (OWASP ASVS)
  • V1.2.1
  • V1.2.3
  • V1.2.5
  • V1.2.8

Mitigating Controls (NIST 800-53 r5) AI

Developer testing and evaluation finds code paths that accept and execute externally influenced strings.

SI-10 directly requires validation of information inputs to reject malformed or special-element content before it reaches downstream parsers.

Least privilege limits the damage an injected code fragment can perform once executed.

Requiring documented secure development standards and tools enforces use of safe code-generation APIs and escaping.

Mitigating Controls (NIST CSF 2.0) AI

Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.

PR.PS-06 mostly match
prevents

Secure SDLC practices directly require input validation and output encoding that prevent injection flaws.

PR.DS-10 none match
prevents

PR.DS-10 protects runtime data confidentiality/integrity but has no bearing on neutralizing externally influenced input during code generation, so neither direction shows any preventive effect.

Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI

Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.

finds

Security testing in development catches injection vulnerabilities before release.

A.8.15 Logging partial match
finds

Logging supports detection of injection attempts but does not prevent the weakness.

finds

Monitoring activities can identify active injection attacks after they occur.

prevents

Secure development life cycle mandates input validation and output encoding that directly prevent injection flaws.

prevents

Application security requirements explicitly call for controls against injection attacks in software design.

prevents

Secure architecture principles reduce injection surfaces but do not prescribe specific neutralization techniques.

References