CVE-2026-69147
vLLM is an inference and serving engine for large language models. Prior to 0.28.0, request bodies for Chat Completions and Responses can se
CVSS
6.5
Medio
EPSS
—
KEV
—
Exploit Today
—
0-100
Publicado: 16 sept 2026 · Última mod.: 16 sept 2026 · CWE-400 · CWE-770
Sin historial EPSS suficiente todavía.
vLLM is an inference and serving engine for large language models. Prior to 0.28.0, request bodies for Chat Completions and Responses can set media_io_kwargs.video.video_backend to pynvvideocodec, and MediaConnector.fetch_video forwards that choice to VideoMediaIO even when startup configuration selected a software decoder. The engine's _reserve_mm_ipc_gpu_memory logic budgets decoder memory only from static configuration, so the request-selected VIDEO_LOADER_REGISTRY backend can create a CUDA context, decoder surfaces, and decoded-frame allocations that were not removed from the engine's KV-cache budget. An attacker able to submit video requests to a video-capable GPU deployment with PyNvVideoCodec installed can exhaust shared GPU memory, causing request failures, worker crashes, or denial of service. The first release containing the fix is version 0.28.0.
- github.comhttps://github.com/vllm-project/vllm/commit/283893c72292ede38d277e3cd2b9b64c3e4f1dda
- github.comhttps://github.com/vllm-project/vllm/commit/ba22152096b2484faa3579624a253d54804d876d
- github.comhttps://github.com/vllm-project/vllm/pull/47259
- github.comhttps://github.com/vllm-project/vllm/security/advisories/GHSA-8pw2-6jv3-mj5j
CVECVSSEPSSKEVRExplotTítuloVis.
CVE-2026-689047.0 ALT—
———node-opcua is an OPC UA implementation for TypeScript and Node.js. From 2.0.0 until 2.170.0, node-opcua clients using the default keepSessionAlive setting can enter a repeated reconnection cycle when an OPC UA server's clock skew causes BadInvalidTimestamp responses. ClientSessionKeepAliveManager._ping_server treated the server-originated ServiceFault as a network outage and forced a transport reconnect, while ClientTCP_transport._on_ACK_response used socket.end() after failed HEL/ACK negotiation and could leave the connection in FIN-WAIT-2 when the peer did not close. Repetition at the keepAliveInterval accumulates file descriptors and memory until the client process or container can be terminated by resource exhaustion. This issue is fixed in version 2.170.0.3hCVE-2026-571736.5 MED—
———vLLM is an inference and serving engine for large language models. Prior to 0.24.0, the input_audio handling path for /v1/chat/completions calls AudioMediaIO.load_bytes or AudioMediaIO.load_file without passing VLLM_MAX_AUDIO_DECODE_DURATION_S to the shared audio decoder. An unauthenticated client can therefore submit a small compressed audio input that expands into a very large float32 PCM allocation, bypassing the duration guard already used by /v1/audio/transcriptions and causing an out-of-memory worker crash. Inline data URLs reach this path without being bounded by VLLM_AUDIO_FETCH_TIMEOUT. The issue affects deployments serving an audio-capable model, and authentication changes only the deployment-specific reachability. This issue is fixed in version 0.24.0.3hCVE-2026-796517.5 ALT—
———A flaw was found in the theme localization endpoints of the keycloak-services component, which is the core service responsible for authentication flows and theme management in Keycloak. The issue occurs because the system accepts arbitrary locale tags from unauthenticated requests and stores them in a permanent in-memory cache without limits. An attacker can exploit this by sending a large number of unique locale tags, eventually causing the server to run out of memory and crash.4hCVE-2026-77409——
———RabbitMQ amqp091-go is a Go AMQP 0.9.1 client. Prior to 1.13.0, Channel.dispatch in channel.go, confirms.confirm in confirms.go, and Connection.dispatch0 in connection.go synchronously send publisher confirmations, flow-control events, consumer cancellations, returned messages, including NotifyConfirm events and connection block notifications, to application-provided channels. If a listener channel is unbuffered, full, or not drained promptly, the sole reader goroutine blocks and stops processing frames, acknowledgments, deliveries, and heartbeats. Broker-driven event bursts can therefore cause connection stalls, missed heartbeats, deadlocks, and disconnection. This issue is fixed in version 1.13.0.5hCVE-2026-77403——
———RabbitMQ amqp091-go is a Go AMQP 0.9.1 client. Prior to 1.13.0, Connection.openTune in connection.go accepts a server-advertised FrameMax below the AMQP frameMinSize value of 4096 bytes because the connection negotiation loop does not enforce the protocol minimum. A malicious or compromised AMQP broker can therefore advertise an extremely small FrameMax, causing later client publications to be fragmented into excessive numbers of frames and write operations. This can consume CPU and stall the client or its host. This issue is fixed in version 1.13.0.4hCVE-2026-631287.5 ALT—
———RMCP is an official Rust SDK for the Model Context Protocol. Prior to 2.0.0, the rmcp crate's stateful Streamable HTTP server in crates/rmcp/src/transport/streamable_http_server/tower.rs allows an unauthenticated client to send a well-formed JSON-RPC POST that is not an initialization request, or an initialization request with a mismatched protocol header, causing StreamableHttpService::handle_post to call LocalSessionManager.create_session before validating the message. An early validation failure returns without removing the inserted LocalSessionHandle from LocalSessionManager.sessions, permanently retaining session and channel state for the server process lifetime. Repeated requests can grow the shared session table without bound, degrade legitimate-client latency through lock contention, exhaust memory, and terminate the server. This issue is fixed in version 2.0.0.4h