CVE-2026-86317: ggml-org llama.cpp RPC Server ggml-rpc.cpp deserialize_tensor assertion
A vulnerability was detected in ggml-org llama.cpp up to 0.4.0. This impacts the function rpcserver::deserializetensor of the file ggml/src/ggml-rpc/ggml-rpc.cpp of the component RPC Server. Performing a manipulation of the argument ne results in reachable assertion. The attack is possible to be carried out remotely. The reported GitHub issue was closed automatically due to inactivity.
Affected Software
Event History
Frequently Asked Questions
Who is exposed to this issue?
Deployments of ggml-org llama.cpp up to version 0.4.0 that expose or use the RPC Server component are affected. The attack can be performed remotely, so network reachability to the RPC Server is the relevant exposure condition.
What does an attacker need to exploit it?
An attacker needs to manipulate the ne argument processed by rpc_server::deserialize_tensor in ggml/src/ggml-rpc/ggml-rpc.cpp. No privileges or user interaction are required according to the supplied vector.
What is the likely impact of successful exploitation?
The manipulation reaches an assertion, producing an availability impact. The provided scoring information indicates no confidentiality or integrity impact and a low availability impact.
How can I determine whether my environment is affected?
Check whether your deployment uses ggml-org llama.cpp version 0.4.0 or earlier and whether its RPC Server component is reachable by remote clients. Review RPC Server handling of tensor deserialization requests involving the ne argument.