NVIDIA is building a security layer to stop them
Needs context
NVIDIA has announced OpenShell, the Open Agent Safety Platform, Sentry (BlueField DPU enforcement) and related tools to enforce policies, sandbox agents, and quarantine misbehaving agents, but these are framed as security, governance and containment measures rather than an effort solely to "stop" AI agents.
This claim may mislead readers without added context.
AI agents are getting powerful enough
Supported
Multiple industry reports and NVIDIA materials describe rapidly increasing agent usage and capability, so the general statement that agents are becoming more capable is supported, though it is imprecise and lacks quantitative detail.