CVE-2024-12055: Ollama: DoS via malicious gguf model file upload

HIGH PoC AVAILABLE CISA: TRACK*

Published March 20, 2025

CISO Take

Any public-facing Ollama instance on version <=0.3.14 can be crashed with a single unauthenticated model upload, taking down all LLM inference on that server. Upgrade to the latest Ollama release immediately and restrict the model upload API to authenticated users only. If running Ollama in shared or production environments, treat this as critical until patched.

What is the risk?

High severity (CVSS 7.5) with network-accessible attack vector, zero authentication required, and trivial exploitation. Ollama is widely deployed in enterprise AI development and self-hosted inference stacks—often with default configurations that expose the API without authentication on internal networks or inadvertently to the internet. While not in CISA KEV and no confirmed active exploitation at time of writing, the near-zero exploitation barrier and broad deployment footprint make this a high-priority patch target. Impact is limited to availability—no confidentiality or integrity compromise—but sustained DoS of inference services can disrupt AI-dependent business processes.

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

7.5 / 10

EPSS

0.8%

chance of exploitation in 30 days

Higher than 53% 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 Network

AC Low

PR None

UI None

S Unchanged

C None

I None

A High

What should I do?

6 steps

PATCH

Upgrade Ollama to a version >0.3.14 immediately—this is the primary fix.
RESTRICT API ACCESS

Place Ollama behind a reverse proxy (nginx/Caddy) with authentication required before any API call reaches the service.
BLOCK MODEL UPLOADS

If model upload/creation is not required, block /api/push and model create endpoints at the network layer or via proxy ACLs.
NETWORK SEGMENTATION

Ollama should never be directly internet-exposed; bind to localhost or internal network interfaces only, enforce VPN access.
DETECT

Monitor for repeated unexpected Ollama process crashes or restarts as a potential exploitation indicator—alert on more than 2 restarts within 10 minutes.
AUDIT

Inventory all Ollama deployments including developer laptops, test environments, and CI/CD pipelines that may run with default insecure configurations.

What does CISA's SSVC say?

Decision Track*

Exploitation poc

Automatable Yes

Technical Impact partial

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

How is it classified?

DoS Supply Chain Inference Model Framework AML.T0011.000 - Unsafe AI Artifacts AML.T0029 - Denial of AI Service AML.T0049 - Exploit Public-Facing Application AML.T0076 - Corrupt AI Model

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.9.2 - AI system availability and resilience

NIST AI RMF

MANAGE 2.2 - Mechanisms to sustain deployment and operation of the AI system

OWASP LLM Top 10

LLM04 - Model Denial of Service

Frequently Asked Questions

What is CVE-2024-12055?

Is CVE-2024-12055 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2024-12055, increasing the risk of exploitation.

How to fix CVE-2024-12055?

1. PATCH: Upgrade Ollama to a version >0.3.14 immediately—this is the primary fix. 2. RESTRICT API ACCESS: Place Ollama behind a reverse proxy (nginx/Caddy) with authentication required before any API call reaches the service. 3. BLOCK MODEL UPLOADS: If model upload/creation is not required, block /api/push and model create endpoints at the network layer or via proxy ACLs. 4. NETWORK SEGMENTATION: Ollama should never be directly internet-exposed; bind to localhost or internal network interfaces only, enforce VPN access. 5. DETECT: Monitor for repeated unexpected Ollama process crashes or restarts as a potential exploitation indicator—alert on more than 2 restarts within 10 minutes. 6. AUDIT: Inventory all Ollama deployments including developer laptops, test environments, and CI/CD pipelines that may run with default insecure configurations.

What systems are affected by CVE-2024-12055?

This vulnerability affects the following AI/ML architecture patterns: LLM inference serving, self-hosted AI infrastructure, model serving, agent frameworks, RAG pipelines.

What is the CVSS score for CVE-2024-12055?

CVE-2024-12055 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.82%.

What is the AI security impact?

Affected AI Architectures

LLM inference servingself-hosted AI infrastructuremodel servingagent frameworksRAG pipelines

MITRE ATLAS Techniques

AML.T0011.000 Unsafe AI Artifacts

AML.T0029 Denial of AI Service

AML.T0049 Exploit Public-Facing Application

AML.T0076 Corrupt AI Model

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.9.2

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM04

What are the technical details?

Original Advisory

A vulnerability in Ollama versions <=0.3.14 allows a malicious user to create a customized gguf model file that can be uploaded to the public Ollama server. When the server processes this malicious model, it crashes, leading to a Denial of Service (DoS) attack. The root cause of the issue is an out-of-bounds read in the gguf.go file.

Exploitation Scenario

An adversary identifies a public-facing or network-accessible Ollama instance via Shodan/Censys scanning for the default port 11434. Requiring zero credentials, they craft a malformed GGUF model file with an invalid header or tensor metadata section designed to trigger the out-of-bounds read in gguf.go during parsing. They upload the file via the standard Ollama API model creation endpoint. When the server processes the malicious GGUF file, the out-of-bounds read causes an immediate process crash. All LLM inference is unavailable for every user and application depending on that endpoint. On container auto-restart, the attacker simply re-uploads the file to sustain the outage indefinitely. The entire attack requires only basic knowledge of the GGUF file format and a single HTTP request.

Weaknesses (CWE)

CWE-125 Out-of-bounds Read

CWE-125 — Out-of-bounds Read: The product reads data past the end, or before the beginning, of the intended buffer.

[Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
[Architecture and Design] Use a language that provides appropriate memory abstractions.

Source: MITRE CWE corpus.