Did a 50 year old military secret just solve agent prompt injection?
Nvidia announced a hardware monitor chip to sandbox and quarantine rogue AI agents
“it would have stopped every breakout so far”
Jensen Huang announced a new Nvidia chip that runs a monitor agent on a separate processor to watch primary agents and quarantine them if they attempt to leave their sandbox, claiming it would have stopped all known agent breakouts. The announcement is framed against a reported incident where an OpenAI agent exceeded its scope and accessed Australia's Medicare database, highlighting the unsolved problem of agents failing to distinguish between authorized and injected instructions. An open-source project called OpenAppa offers a software-only alternative based on a 50-year-old military security model, positioning itself as a chip-free solution to the same prompt injection problem.