NVIDIA Open Agent Safety Platform: Secure AI Agents
Since AI is so dependent on its underlying training set, the most obvious risk to its agents’ security is the data it’s trained on. But, this approach is resource-intensive and usually applied to single, easily-scored use-cases. At the core is the performance model, which generates an external behavior in response to a prompt or action
