New attack lets hackers plant hidden instructions in AI memory with a single prompt
The researchers said the method targets AI agents that store past interactions and reuse them during future tasks, enabling malicious content introduced during an initial…