NVIDIA launches OpenShell and Sentry to contain rogue AI agents
NVIDIA just dropped two new security tools (OpenShell and Sentry) to help stop AI systems from going rogue or getting hacked.
The company says these could have even blocked the recent Hugging Face breach.
The timing is spot on, as labs like OpenAI and Anthropic are also dealing with similar scares.
NVIDIA tools support Arm and Intel
OpenShell uses NVIDIA's own processor features to keep AI agents contained, while Sentry steps in with a separate chip to shut down any agents that try to break out.
Working with Arm Holdings and Intel to ensure the system also works on their central processors, it works across different types of processors.
NVIDIA's Ali Golshan, senior director of AI software at NVIDIA, explained that mathematical formulas help spot when an agent is trying to use workarounds.
CEO Jensen Huang summed it up by saying AI safety should be tackled like car safety: through better engineering, not just more rules.