OpenAI stays out of Nvidia's agent safety platform, TechCrunch reports
More than 100 companies back Nvidia's Open Agent Safety Platform, but OpenAI, Amazon, Google and Apple are absent, and OpenAI says it is working with Nvidia privately.
OpenAI is not a public supporter of Nvidia's Open Agent Safety Platform, TechCrunch reported on Tuesday, although it is privately working with Nvidia on agent security. TechCrunch said more than 100 companies back the effort, including Anthropic, Hugging Face, Arm and Intel, while OpenAI, Amazon, Google and Apple are not among them. No other outlet has confirmed the account.
TechCrunch described the platform as pairing OpenShell, an open-source sandbox, with Nvidia Sentry, proprietary monitoring that runs on BlueField-4 processors to detect and shut down unauthorized agent behavior. It said Nvidia announced the platform on September 29. An OpenAI spokesperson told TechCrunch the company is supportive of Nvidia's work.
TechCrunch reported that OpenAI has contributed privately to OpenShell. It also runs its own Defense Factory consortium, which TechCrunch said includes Anthropic, AWS and Google, and builds its own tools such as the Daybreak cybersecurity model. Hugging Face chief executive Clem Delangue was quoted saying that if OpenAI had run the platform on the agents that attacked Hugging Face, it would have caught them first.
Separately, Nvidia published a developer post on Wednesday about NeMo Relay, an observability layer for agents. Nvidia's documentation says it records agent runs as ordered event logs and step-by-step trajectories, and exports OpenTelemetry spans. The post references version 0.8.3. The paper has not tested it, and the post does not link it directly to the safety platform.