CVE-2026-103270

LightLLM through 1.2.0 mounts reinforcement learning control routes on the public HTTP API without authentication checks. Unauthenticated attackers can call endpoints like /pause_generation, /abort_request, /flush_cache, and /init_weights_update_group to disrupt inference operations and wedge workers on deployments started with --enable_rl.

CVSS 4.0

8.7CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N

CVSS 3.1

7.5CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H

EPSS

0.5% chance of exploitation in the next 30 days; higher than 43% of CVEs (scores of 2026-10-25)

CWE

CWE-306

Published

2026-09-30

Modified

2026-09-30

Record from

the NVD, synced 2026-10-19

References

• https://github.com/ModelTC/LightLLM
• https://github.com/ModelTC/LightLLM/issues/1609
• https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/api_http.py#L505-L507
• https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/api_http_rl.py#L51-L139
• https://www.vulncheck.com/advisories/lightllm-through-1.2.0-missing-authentication-on-rl-control-routes

Check your stackJSON