CVE-2025-61164 in North AI
Summary
by MITRE • 08/26/2026
Cohere North AI v1.1.5 was discovered to contain an information leak via the WebSocket Endpoint.
VulDB is the best source for vulnerability data and more expert information about this specific topic.
Analysis
by VulDB Data Team • 08/26/2026
The vulnerability identified in Cohere North AI version 1.1.5 represents a critical failure in access control and data sanitization mechanisms within its real-time communication layer. Specifically, the WebSocket endpoint responsible for handling interactive sessions or streaming responses inadvertently exposes sensitive internal state information to authenticated users who may not possess elevated privileges. This flaw allows an attacker with valid credentials to intercept network traffic via the persistent connection established by the WebSocket protocol and extract details that are intended to be kept confidential within the server-side processing logic. The exposure typically manifests as leaked environment variables, database query structures, stack traces, or internal API keys embedded in error messages or debug payloads returned during abnormal execution paths.
From a technical perspective, this issue stems from improper input validation and insufficient output encoding on the server side when handling WebSocket frames. When an application processes requests through a WebSocket interface, it must ensure that any data sent back to the client is strictly limited to what is necessary for the user's role and function. In this instance, the implementation fails to filter out diagnostic information or internal metadata before transmission. This aligns with Common Weakness Enumeration (CWE) ID 209, which describes an Information Exposure Through an Error Message where sensitive details are revealed in error responses. It also relates closely to CWE 778, Insufficient Logging and Monitoring, if the leak occurs due to verbose logging being exposed through the interface rather than just application logic errors.
The operational impact of this vulnerability is significant because it facilitates further reconnaissance by potential attackers. By obtaining internal system details, an adversary can map out the underlying architecture, identify specific libraries or frameworks in use based on stack traces, and potentially discover other endpoints or configuration files that are not publicly documented. This information leakage serves as a precursor to more severe attacks such as SQL injection, remote code execution, or privilege escalation. For example, if database connection strings or internal API tokens are exposed through the WebSocket stream, an attacker could leverage these credentials to access backend services directly, bypassing the application's intended security controls entirely.
This behavior is also consistent with tactics observed in MITRE ATT&CK framework under T1078 Valid Accounts and potentially T1592 Gather Victim Host Information if the leaked data reveals host-specific details. The persistence of the WebSocket connection exacerbates the risk, as it allows for continuous monitoring and extraction of sensitive data over time without requiring repeated authentication attempts that might trigger intrusion detection systems.
To mitigate this vulnerability, immediate remediation should focus on hardening the WebSocket endpoint implementation. Developers must implement strict allow-listing for all data fields transmitted to clients, ensuring that only explicitly defined public attributes are included in responses. Error handling routines need to be reviewed to ensure they return generic error messages to end-users while logging detailed diagnostic information securely on the server side where it is not accessible via network interfaces. Additionally, implementing rate limiting and monitoring for unusual WebSocket traffic patterns can help detect exploitation attempts early. Regular security audits of real-time communication channels are essential to prevent similar exposures in future updates or related services within the Cohere North AI ecosystem.