Researchers reveal how hidden instructions embedded in indexed documents can coerce background AI agents into exfiltrating data, exposing a new class of prompt‑injection attacks that sidestep corporate AI safeguards.

Read More