CVE-2025-44779: Ollama arbitrary file deletion

CISO Take

Ollama v0.1.33 allows any local user to delete arbitrary files by sending a crafted request to the /api/pull endpoint—no privileges required. Environments running Ollama for LLM inference, including developer workstations and internal GPU servers, should restrict API access to trusted processes and update to a patched release. The local attack vector limits internet exposure, but file deletion targeting model weights, configs, or security tooling is a credible availability and integrity risk.

What is the risk?

Medium risk overall but elevated for AI development and shared inference environments. The local attack vector (AV:L) prevents direct internet exploitation, yet Ollama is widely deployed on developer workstations and internal servers that often lack network isolation on port 11434. No privileges are required (PR:N), meaning any local user or co-resident process can trigger the attack. User interaction (UI:R) adds a mild barrier—typically bypassed via social engineering or a malicious wrapper script. High availability impact (A:H) combined with trivial exploitability makes this dangerous wherever Ollama runs with broad filesystem access.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
Ollama	pip	—	No patch
174.6K 1.6K dependents Pushed 3d ago 12% patched ~0d to patch Full package profile →

Do you use Ollama? You're affected.

How severe is it?

CVSS 3.1

6.6 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 5% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

○ Public PoC indexed (trickest/cve)

Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AV Local

AC Low

PR None

UI Required

S Unchanged

C Low

I Low

A High

What should I do?

6 steps

Update Ollama beyond v0.1.33—check the official GitHub repository (github.com/ollama/ollama) for the patched release; no confirmed patch version was available in NVD at time of analysis.
Restrict access to the Ollama API (default port 11434) via firewall rules or by binding exclusively to 127.0.0.1; never expose it to untrusted networks.
Run Ollama under a dedicated service account with the minimum filesystem permissions required—model directory only.
Monitor and alert on anomalous requests to the /api/pull endpoint, particularly payloads with path traversal patterns (../, %2F, encoded slashes).
Audit all Ollama deployments in CI/CD pipelines and shared developer environments.
Consider AppArmor or seccomp profiles to restrict the filesystem operations Ollama can perform.

What does CISA's SSVC say?

Decision Track*

Exploitation poc

Automatable No

Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

DoS Data Leakage Inference Framework AML.T0011 - User Execution AML.T0029 - Denial of AI Service AML.T0037 - Data from Local System AML.T0049 - Exploit Public-Facing Application

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.6.2 - AI system security

NIST AI RMF

MANAGE 2.2 - Mechanisms to sustain the value of deployed AI systems

OWASP LLM Top 10

LLM06:2025 - Excessive Agency

Frequently Asked Questions

What is CVE-2025-44779?

Ollama v0.1.33 allows any local user to delete arbitrary files by sending a crafted request to the /api/pull endpoint—no privileges required. Environments running Ollama for LLM inference, including developer workstations and internal GPU servers, should restrict API access to trusted processes and update to a patched release. The local attack vector limits internet exposure, but file deletion targeting model weights, configs, or security tooling is a credible availability and integrity risk.

Is CVE-2025-44779 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2025-44779, increasing the risk of exploitation.

How to fix CVE-2025-44779?

1. Update Ollama beyond v0.1.33—check the official GitHub repository (github.com/ollama/ollama) for the patched release; no confirmed patch version was available in NVD at time of analysis. 2. Restrict access to the Ollama API (default port 11434) via firewall rules or by binding exclusively to 127.0.0.1; never expose it to untrusted networks. 3. Run Ollama under a dedicated service account with the minimum filesystem permissions required—model directory only. 4. Monitor and alert on anomalous requests to the /api/pull endpoint, particularly payloads with path traversal patterns (../, %2F, encoded slashes). 5. Audit all Ollama deployments in CI/CD pipelines and shared developer environments. 6. Consider AppArmor or seccomp profiles to restrict the filesystem operations Ollama can perform.

What systems are affected by CVE-2025-44779?

This vulnerability affects the following AI/ML architecture patterns: local LLM inference, model serving, AI development workstations, internal AI infrastructure, CI/CD model pipelines.

What is the CVSS score for CVE-2025-44779?

CVE-2025-44779 has a CVSS v3.1 base score of 6.6 (MEDIUM). The EPSS exploitation probability is 0.16%.

What is the AI security impact?

Affected AI Architectures

local LLM inferencemodel servingAI development workstationsinternal AI infrastructureCI/CD model pipelines

MITRE ATLAS Techniques

AML.T0011 User Execution

AML.T0029 Denial of AI Service

AML.T0037 Data from Local System

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.6.2

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM06:2025

What are the technical details?

Original Advisory

An issue in Ollama v0.1.33 allows attackers to delete arbitrary files via sending a crafted packet to the endpoint /api/pull.

Exploitation Scenario

An attacker with local access to a machine running Ollama—whether via a compromised developer account, a malicious process sharing the host, or a CSRF-triggered request from the browser—sends a crafted HTTP POST to http://localhost:11434/api/pull with a manipulated model name parameter that traverses the filesystem. Due to CWE-20 (improper input validation) and CWE-552 (unauthorized file access), the Ollama service processes the malformed input and deletes an arbitrary file accessible under its running permissions. In a realistic scenario, an attacker embeds the malicious pull request inside a developer tool or shell script, satisfying the UI:R requirement through social engineering. Target files could include model weights (DoS on inference), Ollama's config (service crash), or SSH authorized_keys (persistence setup for further compromise).

Weaknesses (CWE)

CWE-20 Improper Input Validation CWE-552 Files or Directories Accessible to External Parties

CWE-20 — Improper Input Validation: The product receives input or data, but it does not validate or incorrectly validates that the input has the properties that are required to process the data safely and correctly.

[Architecture and Design] Consider using language-theoretic security (LangSec) techniques that characterize inputs using a formal language and build "recognizers" for that language. This effectively requires parsing to be a distinct layer that effectively enforces a boundary between raw input and internal data representations, instead of allowing parser code to be scattered throughout the program, where it could be subject to errors or inconsistencies that create weaknesses. [REF-1109] [REF-1110] [REF-1111]
[Architecture and Design] Use an input validation framework such as Struts or the OWASP ESAPI Validation API. Note that using a framework does not automatically address all input validation problems; be mindful of weaknesses that could arise from misusing the framework itself (CWE-1173).

Source: MITRE CWE corpus.