Executive brief
Ollama, a tool used to run large language models locally, is vulnerable to a flaw that allows an attacker to crash the service. By sending a specially crafted image through the chat interface, an attacker can cause the application to stop responding to all users. This results in a total service outage until the software is manually restarted.
Technical details
A null pointer dereference vulnerability exists in Ollama's multi-modal image processing via the /api/chat endpoint. The application fails to validate that base64-encoded image data represents valid media before passing it to the mtmd_helper_bitmap_init_from_buf function. This function returns NULL for malformed input, which is subsequently dereferenced without a check, causing a segmentation fault. A remote, unauthenticated attacker can exploit this by sending malformed base64 data to crash the runner process. The vulnerability affects versions 0.11.5-rc0 through 0.13.5.
Affected products
- Ollama Ollama 0.11.5-rc0 through 0.13.5
- Red Hat Red Hat OpenShift AI (RHOAI)
Timeline
- 2026-01-12: disclosed
- 2026-01-12: advisory
References
- https://https//github.com/ollama/ollama
- https://huntr.com/bounties/172df98b-07cd-41ea-a628-366f8cd525c0
- https://ollama.com/
- https://www.vulncheck.com/advisories/ollama-multi-modal-image-processing-null-pointer-dereference
- https://access.redhat.com/security/cve/CVE-2025-15514
- https://bugzilla.redhat.com/show_bug.cgi?id=2428828
- https://security.access.redhat.com/data/csaf/v2/vex/2025/cve-2025-15514.json