CVE-2026-66027 describes a broken access control vulnerability in kortix-ai's suna product prior to version 0.9.102. The flaw exists in the message queue API, where ownership and account isolation checks are missing. Authenticated attackers can leverage this to read or delete other users' sessions, access pending prompt queues, and inject arbitrary prompts into victim sessions. This leads to the victim's AI agent processing malicious messages under the victim's credentials and permissions, potentially compromising the victim's environment.

The vulnerability allows authenticated attackers to bypass access controls and manipulate other users' message queues and sessions. This can result in unauthorized disclosure of pending prompts, deletion of sessions, and injection of malicious prompts that execute with the victim's permissions. The impact is high as it compromises confidentiality, integrity, and availability of user sessions and data within the affected system.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Since no official fix or patch is indicated, users should monitor vendor communications for updates. Until a patch is available, restrict access to authenticated users and consider additional access controls or network segmentation to limit exploitation potential.