---
id: CVE-2026-92983
title: >-
  InternLM LMDeploy through 0.17.0 in DistServe prefill/decode disaggregation
  mode fails to release scheduler sessions because the proxy uses user-facing
  session IDs instead of internal scheduler keys
summary: >-
  InternLM LMDeploy through 0.17.0 in DistServe prefill/decode disaggregation
  mode fails to release scheduler sessions because the proxy uses user-facing
  session IDs instead of internal scheduler keys. Unauthenticated attackers can
  send co…
severity: high
cvss: 7.5
cvssVector: 'CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H'
cwe:
  - CWE-772
vendor: InternLM
product: lmdeploy
affected:
  - lmdeploy <= 0.17.0
published: '2026-09-17'
updated: '2026-09-22'
sourceUpdated: '2026-09-22T20:53:07.383'
source: NVD
sourceUrl: 'https://nvd.nist.gov/vuln/detail/CVE-2026-92983'
references:
  - url: 'https://github.com/InternLM/lmdeploy'
    label: disclosure@vulncheck.com
  - url: >-
      https://github.com/InternLM/lmdeploy/blob/v0.17.0/lmdeploy/pytorch/disagg/conn/engine_conn.py#L93-L99
    label: disclosure@vulncheck.com
  - url: 'https://github.com/InternLM/lmdeploy/issues/4967'
    label: disclosure@vulncheck.com
  - url: >-
      https://www.vulncheck.com/advisories/internlm-lmdeploy-through-0.17.0-memory-exhaustion-via-session-id-mismatch
    label: disclosure@vulncheck.com
  - url: 'https://github.com/InternLM/lmdeploy/issues/4967'
    label: 134c704f-9b21-4f2e-91b3-4a467353bcc0
tags:
  - nvd
  - cve.org
  - exploit-available
epss: 0.00366
epssPercentile: 0.30466
exploitAvailable: true
ssvc:
  exploitation: poc
  automatable: 'yes'
  technicalImpact: partial
  timestamp: '2026-09-18T19:18:23.473849Z'
ingestedAt: '2026-09-17T15:20:39.002Z'
---

## Overview

InternLM LMDeploy through 0.17.0 in DistServe prefill/decode disaggregation mode fails to release scheduler sessions because the proxy uses user-facing session IDs instead of internal scheduler keys. Unauthenticated attackers can send completion requests to the proxy endpoint that accumulate unreleased scheduler metadata and memory until the prefill worker is out-of-memory killed.

## Remediation

Refer to the linked advisories for vendor-supplied fixes and affected version ranges.