Fake AI Agent Skill Passed Security Scans and Reportedly Reached 26,000 Agents

Summary: Security firm AIR built a fake AI agent skill named brand-landingpage, pushed it through a skill marketplace and an Instagram ad, and says it reached roughly 26,000 agents, including some on corporate accounts, while every skill scanner it tested marked the skill safe. The skill pointed agents to an external page on a domain AIR controls, initially showing genuine Google Stitch documentation to pass review, then swapped after wide installation to instruct the agent to download and run a script that mailed the user's email address back to AIR.
Key Takeaway: The scan happens once, but the page a skill points an agent to can be rewritten at any time afterward, so a clean verdict at install says nothing about what the skill does next. AIR's 26,000 figure is its own and unconfirmed, but the external-link blind spot it exploited was independently demonstrated by Trail of Bits three weeks earlier against ClawHub's detector, Cisco's scanner, and all three scanners wired into skills.sh.

Read the original at The Hacker News ↗