/

CVE-2026-85695

Severity:

CRITICAL RISK

(9.4)

Authentication Bypass (CWE-306) in FastChat Workers

CVE-2026-85695

Deferred

Vulnerability type

Category:

Authentication Bypass

CWE:

CWE-306

Impact:

Server-Side Request Forgery and Sensitive Data Interception

Severity:

CRITICAL RISK

(9.4)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

FastChat has an authentication bypass in the /register_worker endpoint that may allow unauthenticated users to register arbitrary worker addresses under existing model names. This can enable server-side request forgery, interception of prompts and responses, and probing of internal services when the endpoint is reachable and worker registration is not otherwise protected.

Affected context

The issue occurs in FastChat deployments that expose the worker-registration interface without effective authentication or authorization. It becomes risky when the service accepts attacker-controlled worker addresses and routes model traffic based on worker or model identifiers, allowing a malicious registration to influence requests within the worker mesh.

Exploit preconditions

  • Attack surface: The FastChat /register_worker endpoint
  • Required access: Network reachability to the endpoint; valid credentials may not be required
  • Trigger point: Submission of a worker registration associated with a victim model name
  • Why exploitation is possible: The endpoint may trust unauthenticated registration data and allow attacker-controlled worker addresses to participate in request routing

Why this happens

The underlying flaw is missing authentication and authorization on a sensitive administrative registration operation. This maps to CWE-306: Missing Authentication for Critical Function, because the application permits an unauthenticated party to influence worker topology and model routing. When registration data is accepted without establishing trust in the caller or validating the destination, the same weakness can facilitate server-side requests to attacker-selected or internal addresses.

Potential impact

  • An attacker may intercept user prompts, uploaded images, and model responses
  • The service could send server-side requests to attacker-controlled or internal destinations
  • Internal network ports and services could be probed
  • Model availability may be disrupted through malicious worker registration
  • Sensitive information could be exposed through manipulated worker traffic

General mitigation guidance

Mitigation generally focuses on protecting worker registration as a privileged control-plane operation.

Common defensive patterns include:

  • Require strong authentication and explicit authorization for worker registration
  • Use secure-by-default configurations that do not expose registration endpoints publicly
  • Bind registrations to trusted identities, attestations, or an approved worker inventory
  • Validate model identifiers and worker addresses against strict allowlists
  • Restrict outbound network access from workers and control-plane services
  • Apply least-privilege permissions and isolate worker traffic from sensitive internal networks
  • Avoid sending sensitive user content to workers whose identity or destination has not been verified

How teams detect issues like this

Organizations can look for this issue through:

  • Static and dynamic analysis of registration endpoints for missing authentication middleware
  • Code review checks for routes that modify worker state without authorization decisions
  • Monitoring for unauthenticated registration attempts or unexpected worker addresses
  • Alerts when a model name is associated with new, unapproved, or rapidly changing workers
  • Network policy enforcement and egress monitoring for unexpected internal-port probes
  • Automated security review tools that identify unauthenticated administrative endpoints and SSRF-prone URL handling

Sources

SourceLink
github.comgithub.com/lm-sys/FastChat
github.comgithub.com/lm-sys/FastChat/blob/v0.2.36/fastchat/serve/controller.py
github.comgithub.com/lm-sys/FastChat/issues/3886
vulncheck.comvulncheck.com/advisories/fastchat-unauthenticated-worker-registration-ssrf-and-model-spoofing

Frequently asked

questions

It may be remotely exploitable when the /register_worker endpoint is network-reachable and does not require effective authentication. The practical scope depends on deployment exposure and network controls.

The described bypass does not require normal application credentials. An attacker generally needs connectivity to the registration endpoint and the ability to submit worker registration data.

Protect worker registration with strong authentication, explicit authorization, trusted worker identity validation, and strict address allowlisting. Restrict endpoint exposure and outbound connectivity as additional layers.

Treat worker registration as a privileged control-plane function, review all state-changing endpoints for authentication coverage, monitor topology changes, and enforce network policies that limit unexpected server-side requests.

Find out what's already
exploitable in your codebase.