Vulnerability  ·  2026-08-15

MindsDB Minds Platform unauthenticated RCE via Anton agent prompt injection

VulnerabilityHigh impactGlobalCVE-2026-73678
NVD published CVE-2026-73678 (CVSS 10.0, Critical) on 2026-08-14, describing an unauthenticated remote code execution vulnerability in MindsDB's Minds Platform where crafted prompts submitted to the unprotected /api/v1/responses/ endpoint reach the Anton agent's execution logic, allowing arbitrary OS command execution.
This is a maximum-severity (CVSS 10.0) unauthenticated RCE in an AI agent orchestration platform — an attacker needs no credentials to fully compromise the host running the agent, exposing any connected databases, secrets, or downstream systems the agent has access to.
Unauthenticated attacker submits a crafted prompt to the unprotected POST /api/v1/responses/ endpoint, which reaches the Anton agent's code-execution path and allows arbitrary OS command execution with no authentication required
MindsDB Minds Platform ≤ 26.1.0 (Anton self-improving agent component)
Upgrade MindsDB Minds Platform beyond 26.1.0; restrict network exposure of the /api/v1/responses/ endpoint until patched
NVD CVE-2026-73678MindsDB Minds Platform GitHub
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →