On 28 September 2026 NVIDIA announced an Open Agent Safety Platform: open software and a reference system to set boundaries for agents from test to production, across software, compute, and robots.
The press list is long — Anthropic, Microsoft, CrowdStrike, Hugging Face, and many others. That is a coalition, not a measurement. What to watch is whether a named team can fail an agent in a test and stop it in production.
Safer agents need a kill switch and a person who owns the action. A platform name does not replace that.
OriginalNVIDIAinvestor.nvidia.comRead the full story