Researchers showed AI agents can be manipulated by hidden prompts to leak sensitive files, highlighting a growing security ...