AIR
A research organization in the AI safety domain that focuses on identifying and disclosing security vulnerabilities within AI agent ecosystems.
Timeline (last 90 days)
A security company, AIR, built an AI agent skill disguised as a tool from a well-known company and successfully passed review to be listed on a popular marketplace.
All scanners that examined the malicious skill, including those from well-known security vendors, reported it as clean upon inspection, because it genuinely was clean at the time of inspection.
The skill referenced a documentation link that served real documentation at scan time; after enough users had installed it, the researchers changed the link's content to instruct the agent to download and execute a script.
All Facts (3)
All scanners that examined the malicious skill, including those from well-known security vendors, reported it as clean upon inspection, because it genuinely was clean at the time of inspection.
50%UnverifiedThe skill referenced a documentation link that served real documentation at scan time; after enough users had installed it, the researchers changed the link's content to instruct the agent to download and execute a script.
50%UnverifiedA security company, AIR, built an AI agent skill disguised as a tool from a well-known company and successfully passed review to be listed on a popular marketplace.
50%