CVE-2026-111160

7.5High

Published 15 Oct 2026 · NVD status: Deferred · AI triage by gemini-3.8-flash

7.5High

Attack vector: Network
Complexity: Low
Privileges needed: None
User interaction: None
Impact: High impact on availability
Weakness: CWE-502

Impact

Unauthenticated remote attackers can permanently hang all inference requests by crashing the worker thread with a single HTTP request. They only need network access to the /generate endpoint.

Root cause

The /generate endpoint allows user-controlled input into the internal pd_kv_trans_params field, triggering an unhandled pickle.loads failure during InferSamplingParams deserialization.

Affected systems

LightLLM through 1.2.0

CVSS v3.1 · CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H

Fix & remediation

Upgrade to a version newer than 1.2.0 once available. In the meantime, block external access to the /generate endpoint or use a reverse proxy/WAF rule to strip the pd_kv_trans_params field from incoming request bodies.

shell

$ pip list | grep lightllm

AI-suggested. Run it in staging first and check it against the vendor advisory. Why?

Ad · In-content banner

NVD description

LightLLM through 1.2.0 contains a denial of service vulnerability that allows unauthenticated attackers to crash the inference worker by supplying the internal pd_kv_trans_params field to the /generate endpoint. Attackers can send a single request whose parameters cause pickle.loads to fail in InferSamplingParams, killing the inference thread while /health stays up and all later requests hang until restart.

References

Source: nvd.nist.gov/vuln/detail/CVE-2026-111160