{"id":"CVE-2026-119052","kind":"vulnerability","aliases":["CVE-2026-119052"],"related":[],"title":"llama.cpp RPC server: oversized message crashes the server for all clients","description":"llama.cpp b2877 before b11935 contains an uncaught exception vulnerability in the ggml-rpc server that allows unauthenticated remote attackers to crash the rpc-server by sending oversized message lengths. After the HELLO handshake, attackers can send a command such as SET_TENSOR with a length above std::vector::max_size(), triggering an uncaught std::length_error that aborts the process and denies service to all clients.","summary":"llama.cpp's ggml-rpc server fails to catch an exception when a client sends a message with an oversized length field after the HELLO handshake. An unauthenticated remote attacker can crash the rpc-server process and deny service to every connected client.","severity":"high","cvss":8.7,"epss":null,"exploited":false,"exploited_at":null,"products":[],"ecosystems":[],"tags":["ai-inference","ai-tooling","ggml-rpc","llama.cpp","self-hosted"],"not_affected_if":["rpc-port-not-exposed","rpc-server-not-running"],"how_to_check":"- Check the llama.cpp build version the rpc-server runs, for example with `rpc-server --version` or the build tag in your deployment; affected builds are from `b2877` up to but not including `b11935`.\n- Check whether the rpc-server listens on a network interface reachable from untrusted hosts (the RPC port is open to the network).","fix":"- llama.cpp: move to `b11935` or later.","status":"open","published_at":"2026-10-29T14:17:52+00:00","first_seen_at":"2026-10-29T14:25:23+00:00","triaged_at":"2026-10-29T14:25:51+00:00","updated_at":"2026-10-29T14:25:51+00:00","category":"ai-tooling","confidence":0.8,"triage":{"at":"2026-10-29T14:25:51Z","model":"claude-haiku-5-5","output":{"fix":"- llama.cpp: move to `b11935` or later.","reach":"network","title":"llama.cpp RPC server: oversized message crashes the server for all clients","impact":"medium","summary":"llama.cpp's ggml-rpc server fails to catch an exception when a client sends a message with an oversized length field after the HELLO handshake. An unauthenticated remote attacker can crash the rpc-server process and deny service to every connected client.","audience":"developers","category":"ai-tooling","nicknames":[],"who_cares":["llama.cpp","ggml-rpc","self-hosted","ai-inference"],"confidence":0.8,"fix_status":"available","needs_auth":"no","worst_case":"remote attacker crashes the rpc-server, denying service to all clients","analyst_note":"","how_to_check":"- Check the llama.cpp build version the rpc-server runs, for example with `rpc-server --version` or the build tag in your deployment; affected builds are from `b2877` up to but not including `b11935`.\n- Check whether the rpc-server listens on a network interface reachable from untrusted hosts (the RPC port is open to the network).","plain_action":"","needs_analyst":false,"not_affected_if":["rpc-server-not-running","rpc-port-not-exposed"],"who_is_affected":"Teams that run llama.cpp's ggml-rpc server (rpc-server) to share model compute over the network are exposed. Local-only use of llama.cpp without the RPC server is not affected by this flaw as described. Teams that keep the RPC port off untrusted networks have less exposure.","user_interaction":"no","needs_special_setup":"yes","severity_if_unknown":"high"},"cost_usd":0.000706,"input_tokens":4172,"output_tokens":577,"severity_before":"high"},"analyst_status":null,"analyst_note":null,"search":"'-119052':14A,17A '-2026':13A,16A 'abort':134C 'ai':19B,22B 'ai-infer':18B 'ai-tool':21B 'allow':91C 'attack':60B,94C,110C 'b11935':78C 'b2877':76C 'catch':39B 'client':11A,44B,74B,142C 'command':114C 'connect':73B 'contain':79C 'crash':6A,62B,96C 'cve':12A,15A 'deni':69B,138C 'error':132C 'everi':72B 'except':41B,82C 'fail':37B 'field':52B 'ggml':25B,34B,87C 'ggml-rpc':24B,33B,86C 'handshak':56B,109C 'hello':55B,108C 'host':30B 'infer':20B 'length':51B,105C,121C,131C 'llama.cpp':1A,27B,31B,75C 'max':125C 'messag':5A,47B,104C 'overs':4A,50B,103C 'process':67B,136C 'remot':59B,93C 'rpc':2A,26B,35B,65B,88C,99C 'rpc-server':64B,98C 'self':29B 'self-host':28B 'send':45B,102C,112C 'server':3A,8A,36B,66B,89C,100C 'servic':70B,139C 'set':117C 'size':126C 'std':123C,130C 'tensor':118C 'tool':23B 'trigger':127C 'unauthent':58B,92C 'uncaught':81C,129C 'vector':124C 'vulner':83C","nicknames":[],"mitigations":[],"who_affected":"Teams that run llama.cpp's ggml-rpc server (rpc-server) to share model compute over the network are exposed. Local-only use of llama.cpp without the RPC server is not affected by this flaw as described. Teams that keep the RPC port off untrusted networks have less exposure.","tier":"B","verified":{},"analyst_revisit_at":null,"fix_status":"available","audience":"developers","plain_action":null,"first_public_at":"2026-10-29T14:17:52+00:00","kev":null,"cvss_vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X","cvss_by":"disclosure@vulncheck.com","poc_count":0,"sources":[{"source":"nvd","external_id":"CVE-2026-119052","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-119052","severity":null,"published_at":"2026-10-29T14:17:52+00:00","modified_at":"2026-10-29T14:17:52+00:00","first_seen_at":"2026-10-29T14:25:23+00:00","fetched_at":"2026-10-29T14:25:23+00:00","note":null}],"documents":[],"notice":"Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai"}