Executive brief
Ollama, a platform used for running large language models locally, is vulnerable to a flaw that allows remote attackers to crash the service. By sending specially crafted data, an attacker can trigger a memory error that shuts down the application, leading to a denial-of-service. This attack does not require a password or any user interaction, potentially disrupting AI-driven operations and services relying on the Ollama infrastructure.
Technical details
A denial-of-service vulnerability exists in Ollama due to improper validation of an array index within the downloadBlob function (CWE-129). The flaw is triggered when the application processes user-supplied data that lacks adequate bounds checking, leading to an out-of-bounds memory access. A remote, unauthenticated attacker can exploit this by sending a malicious request to the affected service, causing the process to crash. As of the advisory date, no official patch was confirmed, and the recommended mitigation is to restrict network access to the Ollama API.
Affected products
- Ollama Ollama 0.7.1
Timeline
- 2025-05-28: disclosed: Vulnerability reported to vendor by ZDI
- 2026-07-08: advisory: Coordinated public release of advisory by ZDI
- 2026-07-13: other: NVD publication date