Vulnerability  ·  2026-07-26

Suna Agentic Platform — Broken Access Control in Message Queue API Exposes Cross-User Prompt Data (CVE-2026-66027)

VulnerabilityHigh impactGlobalCVE-2026-66027
NVD published CVE-2026-66027 (CVSS 8.3, High) on 2026-07-24, describing a broken access control vulnerability in Suna's message queue API allowing authenticated attackers to access and manipulate other users' queued prompts and agent tasks.
In a multi-tenant agentic AI platform, prompt-queue cross-tenant access is a direct data-confidentiality and integrity breach — attackers can read other users' potentially sensitive prompts (which may contain proprietary data or credentials) and tamper with queued agent actions, effectively enabling cross-tenant prompt injection or data exfiltration in production agent deployments.
Missing ownership and account isolation checks in the message queue API allow any authenticated attacker to read and manipulate queue resources belonging to other users, including reading pending prompt queues and manipulating queued agent tasks belonging to other tenants.
Suna (agentic AI platform) before 0.9.102
Upgrade to Suna 0.9.102 or later, which adds proper ownership/account isolation checks on the message queue API.
NVDTheHackerWire
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →