ATTACKERS CAN TURN AN AI AGENT'S OWN TOOLS AGAINST IT

Summary from the TLDR AI newsletter

Goal hijacking works when an agent treats retrieved content as instructions, letting a webpage, email, or document redirect its connected tools.