Live Exposure Defense: From CVE to Confirmed Exposure in 12 Hours – See more

Back
New CVE Detected

CVE-2026-92983 – Unauthenticated Memory Exhaustion / DoS – InternLM LMDeploy ≤ 0.17.0 (DistServe Mod

Be the first to know when new zero-days emerge:

Subscribe to Email Alerts
Subscribe to RSS Feed

Summary

CVE-2026-92983 is a resource-exhaustion vulnerability (CWE-772: Missing Release of Resource after Effective Lifetime) affecting InternLM’s LMDeploy LLM inference toolkit when deployed with DistServe prefill/decode disaggregation. An unauthenticated attacker can send ordinary completion requests to the DistServe proxy that cause the prefill engine to accumulate unreleased scheduler metadata, eventually exhausting memory and crashing the prefill worker. The flaw carries a CVSS v4.0 base score of 8.7 (High), reflecting its network-based, no-privilege, no-user-interaction denial-of-service impact.

Technical details

  • Root cause: The DistServe proxy tracks sessions using the user-facing/response session ID (remote_session_id) instead of the internal scheduler key that actually owns the preserved sequence/cache metadata on the prefill engine.
  • Trigger conditions: Standard traffic to the proxy’s /v1/completions endpoint in DistServe (prefill/decode disaggregation) mode is sufficient — no malformed input or authentication is required. Each request is split into a prefill and a decode phase; when the decode side later issues a "free cache" request using the mismatched session ID, the prefill engine logs an "invalid free" and never releases the associated metadata.
  • Attack vector: Network (AV:N), low attack complexity, no privileges or user interaction required.
  • Impact: Sustained/repeated requests cause unbounded growth of prefill-side scheduler sessions, sequences, and seq_map entries, driving continuous RSS growth until the prefill engine process is killed by an out-of-memory condition. The DistServe container may appear to be running while it is unresponsive, and health checks can begin failing (HTTP 503), leading to proxy node loss and service degradation/denial of service for all users of the deployment.

Affected software

  • InternLM LMDeploy (PyPI package lmdeploy) — all versions through 0.17.0 when running in DistServe (prefill/decode disaggregation) mode.
  • Standard (non-disaggregated) LMDeploy deployments that do not use DistServe are not affected by this specific issue.

Severity

  • CVSS v4.0: 8.7 (High) — 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
  • CVSS v3.1 (equivalent): 7.5 (High) — CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H

Mitigation and recommended actions

  • Immediate: As of this writing, no patched LMDeploy release addressing this specific DistServe session-tracking flaw has been published (the latest available release is 0.17.0, which remains vulnerable). Monitor the InternLM LMDeploy GitHub repository and upgrade to a fixed release as soon as one is issued.
  • If no patch is available:
    • Restrict network access to the DistServe proxy and its /v1/completions endpoint to trusted clients only (e.g., via network segmentation, firewall rules, or an authenticating reverse proxy) rather than exposing it directly to the internet.
    • Monitor prefill worker memory (RSS) usage and configure alerting/auto-restart on abnormal growth to reduce the impact of an exhaustion event.
    • Consider disabling DistServe prefill/decode disaggregation mode where it is not operationally required, reverting to standard serving mode until a fix is available.
    • Apply rate limiting on completion requests reaching the proxy to slow the rate of metadata accumulation.

References

Are you exposed?

Get a free report of your organization’s exposure to this CVE and threat

  • Mapping of all Assets with this Technology
  • Identification of potentially exposed assets to this CVE
  • Confirmation of verified exploitable assets
Get Exposure Report

How IONIX’s External Exposure Management Platform Detects and Validates
Zero-Days to Shrink MTTR

1

Map your entire attack surface (continously)

IONIX uses multi-factor discovery methods, including DNS analysis, certificate mapping, metadata inspection, and more, to automatically map every internet-facing asset across your environment. This includes cloud instances, third-party platforms, shadow IT, and even forgotten infrastructure that traditional tools miss.

2

Monitor for new CVEs

Dozens of threat intel feeds using agentic technology are continuously analyzed to detect the appearance of proof-of-concept code, exploit kits, and indicators of active targeting. IONIX goes further by applying AI to proactively evaluate whether emerging vulnerabilities are likely to be exploited, even before PoCs go public.

3

Identify Potential External Exposures

Not all CVEs matter. IONIX filters vulnerabilities by asking attacker-centric questions: Can it be reached from the internet? Does it require authentication? Is it being exploited in the wild? This dramatically reduces noise and focuses teams on threats that can actually be weaponized.

4

Create Safe, Scalable Exploit Validations

IONIX transforms real-world PoCs into safe, non-intrusive test payloads that can be run in production environments without disruption. These simulations are precisely targeted to the systems that are vulnerable, ensuring rapid validation without unnecessary load.

5

Execute Exploit Validations

By combining context about software stack, versioning, exposure status, and reachability, IONIX ensures that only the right payloads are executed against the right assets, maximizing efficiency and minimizing risk.

6

Drive Fast and Actionable Remediation

Results are routed through integrations with ticketing, SOAR, and SIEM tools. Issues are written in plain language, bundled into remediation clusters, and prioritized based on asset criticality, exploitability, and blast radius. This shortens mean time to remediation (MTTR) and empowers teams to act with confidence.

Are you exposed?

Get a free report of your organization’s exposure to this CVE and threat

  • Mapping of all Assets with this Technology
  • Identification of potentially exposed assets to this CVE
  • Confirmation of verified exploitable assets
Get Exposure Report
Close

Subscribe to Threat Center RSS

Copy/paste the link below into your preferred RSS reader or follow these instructions to subscribe to Slack alerts.

Open RSS feedSlack setup instructions

Close

Get Real-Time CVE Alerts to Your Email

Be the first to know when new zero-days emerge