A high-severity information disclosure vulnerability in ash_ai allows authenticated users to capture sensitive API credentials via crafted error messages.

This vulnerability involves the generation of error messages containing sensitive information (CWE-209). By submitting malformed or oversized content that triggers an embedding provider failure, an authenticated user can force the application to return raw error terms, including outbound Authorization headers and API keys, directly to the caller.

The exposure of API keys and provider request details poses a significant risk to organizational security, as it grants attackers the potential to hijack third-party AI services or perform unauthorized actions on behalf of the application. With a CVSS score of 7.1, this vulnerability is considered High, as it enables the compromise of sensitive authentication material that could lead to further downstream attacks or service abuse.

Immediate Action: Update the ash_ai dependency to version 1.0.0 or later, which implements sanitized error logging to prevent the leakage of raw provider error terms.

Proactive Monitoring: Review application logs for unusual spikes in validation errors or repeated attempts to submit malformed input that triggers embedding processes.

Compensating Controls: If immediate patching is not feasible, restrict access to the affected endpoints and ensure that API keys used by the embedding service have the minimum necessary scope and are regularly rotated.

Public Exploit Available: Unknown

Given the ease of exploitation and the severity of credential exposure, organizations should prioritize the update to version 1.0.0. Failure to patch allows authenticated users to potentially exfiltrate production API credentials, creating a substantial risk to both the integrity of the application and the security of third-party AI integrations.