How a malicious AI agent skill passed security checks and reached 26,000 users

AIR says static scanning failed to detect a skill that redirected to a controlled domain and later altered its payload.

Published: 2026-06-24