Database/AI/ML frameworks & serving

llama.cpp: integer overflow parsing audio multimodal projector GGUF files allows code execution
Impact
When llama.cpp parses an audio multimodal projector (mmproj) GGUF file, it does not validate a size, so an integer overflow happens before a buffer is allocated. The ZDI advisory says this lets a crafted file run code in the process that loads it. On a shared inference node, anyone who can get a model or projector file loaded into llama.cpp or llama-server gets code execution as that service, along with its GPU access and whatever credentials or model weights it can read.
Who can reach it
Anyone who can supply a malicious mmproj GGUF file that the target then loads (user interaction: the file must be opened). Pipelines that pull models from untrusted sources or let tenants upload model files are directly exposed.
What to do
Rebuild or update llama.cpp to a version that includes commit e2e7a9b2d04e0404131e551243d4bd399ccfe606, then restart the llama.cpp or llama-server processes. The record names no fixed release tag. Until then, load multimodal projector files only from trusted sources.
References
Related entries
- NVIDIA NemoClaw: code injection in the migration command gives a local user execution as the tool's privilegesCVE-2026-65082 · NVIDIA NemoClaw for Linux (migration command)High
- llama.cpp: crafted GGUF tensor offsets read memory outside the mapped model fileCVE-2024-58399 · llama.cpp llama_model_loader (GGUF tensor offsets)Medium
- vLLM: unbounded cache_salt stalls the EngineCore scheduler thread, denying service to all requestsCVE-2026-100647 · vLLM (OpenAI- and Anthropic-compatible API server, cache_salt handling)Medium
- vLLM: audio clip size limit not enforced, letting unauthenticated clients exhaust node memory and CPUCVE-2026-100648 · vLLM (multimodal chat audio decoding, VLLM_MAX_AUDIO_CLIP_FILESIZE_MB)Medium
- Quarkus LangChain4j: unauthenticated CONNECT frames leak chat scopes until the JVM exitsCVE-2026-108748 · Quarkus LangChain4j chat-scopes WebSocket endpoint (/_chat/routes)Medium
- Crawl4AI: robots.txt fetch bypasses egress controls, giving blind SSRF to metadata endpointsCVE-2026-109239 · Crawl4AI (RobotsParser.can_fetch in the Docker server)Medium
This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.