# CVE-2026-103270: LightLLM authentication bypass in reinforcement learning control API

Source: https://junglewise.ai/threats/cve-2026-103270-lightllm-through-1-2-0-mounts-reinforcement-learning-control  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-09-30. Updated: 2026-09-30  
License: CC BY 4.0 with a link back

CVE-2026-103270 · Severity: high · CVSS 7.5 · Published 2026-09-30

LightLLM is a Python framework that serves large language models for inference. When started with reinforcement learning mode enabled (--enable_rl), the service exposes unauthenticated control endpoints on its public HTTP API that allow attackers to pause generation, abort requests, flush caches, and wedge worker processes, causing service unavailability. Even when RL mode is disabled, these endpoints remain mounted and respond with errors instead of 404, contradicting documented behavior.

Technologies: [ModelTC LightLLM](https://junglewise.ai/threats/technologies/lightllm). Vendors: [ModelTC](https://junglewise.ai/threats/vendors/modeltc).

## Quick answers

**Is CVE-2026-103270 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-103270 being exploited in the wild.

**How severe is CVE-2026-103270?** High severity, with a CVSS score of 7.5, published 30 September 2026.

**Which products and versions are affected?** ModelTC LightLLM through 1.2.0.

**Is there a fix for CVE-2026-103270?** No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.

**How is CVE-2026-103270 attacked?** It is attacked over the network, in a software library.

**Am I affected by CVE-2026-103270?** Check whether you run ModelTC LightLLM at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## Technical details

The vulnerability is a missing authentication check (CWE-306) in the HTTP API router. Reinforcement learning control routes defined in api_http_rl.py are unconditionally included in the public API without credential validation, and an unauthenticated network attacker can invoke endpoints like /pause_generation, /abort_request, /flush_cache, and /init_weights_update_group. The vulnerability requires only network access to the inference HTTP port; with --enable_rl enabled, the endpoints function fully; without it, they incorrectly return 500/417 errors instead of 404, revealing the routes exist.

## Affected products

- ModelTC LightLLM through 1.2.0

## Timeline

- 2026-09-30: disclosed

## References

- [https://github.com/ModelTC/LightLLM](https://github.com/ModelTC/LightLLM)
- [https://github.com/ModelTC/LightLLM/issues/1609](https://github.com/ModelTC/LightLLM/issues/1609)

## Related threats

- [LightLLM denial of service via pd_kv_trans_params unpickling](https://junglewise.ai/threats/cve-2026-111160-lightllm-through-1-2-0-contains-a-denial-of-service)
- [LightLLM visual_only unauthenticated RCE via RPyC pickle deserialization](https://junglewise.ai/threats/cve-2026-103395-lightllm-through-1-2-0-visual-only-deployments-expose-an)
- [LightLLM server-side request forgery in multimodal endpoints](https://junglewise.ai/threats/cve-2026-103243-lightllm-through-1-2-0-fails-to-validate-image-url-and-audio-url)
- [LightLLM memory exhaustion in NCCL control channel](https://junglewise.ai/threats/cve-2026-103042-lightllm-through-1-2-0-contains-a-memory-exhaustion)
- [LightLLM multimodal pickle deserialization in RPyC cache service](https://junglewise.ai/threats/cve-2026-103041-lightllm-through-1-2-0-multimodal-deployments-expose-an)