Skip to main content

Improper Access Control

CVE-2026-44556

Low
0/10

Summary

The /responses endpoint in the OpenAI router accepts any authenticated user and forwards requests directly to upstream LLM providers without enforcing per-model access control. While the primary chat completion endpoint (generate_chat_completion) checks model ownership, group membership, and AccessGrants before allowing a request, the /responses proxy only validates that the user has a valid session via get_verified_user. The affected versions are prior to 0.9.0.

  • LOW
  • NETWORK
  • NONE
  • UNCHANGED
  • NONE
  • LOW
  • LOW
  • HIGH

CWE-284 - Improper Access Control

Listed 5th in the 'OWASP Top Ten', improper (or broken) access control attacks are a fundamental type of vulnerability. This includes a broad range of design flaws that enable users to act outside of their intended permissions. They can use these privileges to gain access to restricted files and functionality such as accessing restricted information, falsifying records, destroying data, or executing commands.

References

Advisory Timeline

  • Published