Description

Cohere North AI v1.1.5 was discovered to contain excessively permissive cross-domain policy with untrusted domains. This occurs via the server failing to validate the Origin header of incoming connection requests.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "9.8",
  "pubDate": "2026-08-26T19:16:44.290Z",
  "pubdate": "2026-08-26T19:16:44.290Z",
  "executiveSummary": "Cohere North AI v1.1.5 suffers from a critical cross-domain policy vulnerability stemming from inadequate Origin header validation.\nThis flaw allows untrusted domains to interact with the application, effectively bypassing cross-origin resource sharing (CORS) security boundaries.\nThe vulnerability is characterized by a failure in the server-side logic to verify the trustworthiness of incoming connection requests, permitting unauthorized third-party origins to interface with the service.\nThe potential impact includes unauthorized data access, the performance of actions on behalf of authenticated users, and the exfiltration of sensitive information processed by the AI system.\nAn attacker can exploit this condition by forcing a victim's browser to execute requests against the target domain, leveraging the lack of strict origin enforcement to read responses or execute state-changing operations.\nThe security risk is high, as it compromises the confidentiality and integrity of cross-domain communication, making the system susceptible to Cross-Site Request Forgery (CSRF) and Cross-Site Scripting (XSS)-related data theft techniques without requiring any specific authentication from the attacker's perspective.",
  "technicalDetails": "The vulnerability resides in the request processing pipeline of Cohere North AI v1.1.5, specifically where the server evaluates cross-domain connection headers.\nThe root cause is an improper implementation of origin-based access control, where the server fails to implement a strict allow-list for incoming requests. Instead of performing a cryptographic or regex-based validation against known-good domains, the server implicitly trusts the provided Origin header or fails to perform a validation check entirely, reflecting the supplied origin back in the Access-Control-Allow-Origin response header.\nExploitation is achieved through a malicious cross-domain request. An attacker hosts a payload on an external site that the victim visits. The malicious script initiates an asynchronous request (via XMLHttpRequest or Fetch API) directed at the Cohere North AI endpoint.\nBecause the server does not enforce origin policy, it accepts the request and includes headers permitting cross-origin access. This allows the attacker's script to read the response body returned by the server, potentially exposing sensitive data, session tokens, or internal API responses.\nThe attack flow follows a sequential pattern: first, the attacker lures an authenticated user to a controlled domain; second, the attacker's malicious script triggers an authenticated request to the vulnerable endpoint; third, the server processes the request despite the untrusted origin; finally, the server responds with a permissive policy, granting the attacker read access to the sensitive response payload.\nThis vulnerability effectively bypasses the browser's Same-Origin Policy (SOP). The exposure is inherent to any network-accessible instance of the product, as it relies on the server-side failure to validate the request source. No specific privilege levels are required on the part of the attacker, as they are piggybacking on the session context of an authorized user.\nPost-exploitation, the attacker may perform unauthorized actions within the scope of the victim’s session, bypass CSRF protections if they are tied to origin checking, or steal proprietary AI inference data, significantly impacting the application's overall security posture."
}