SDSignal Desk

Nvidia's 100-company rogue-agent coalition lacks OpenAI, Google, Amazon and Apple, though OpenAI uses its tools

· Updated · Signal Desk newsroom

Analysis of reporting by TechCrunch. How we work

Image: TechCrunch

TechCrunch notes that Nvidia's Open Agent Safety Platform, backed by more than 100 companies including Anthropic, does not include OpenAI, Amazon, Google or Apple. OpenAI says it supports the work and already uses Nvidia's OpenShell, while pursuing its own safeguards and information-sharing efforts.

Key takeaways

  • More than 100 companies back Nvidia's Open Agent Safety Platform; Anthropic is a supporter.
  • OpenAI, Amazon, Google and Apple are absent from the public list.
  • An OpenAI spokesperson said the company supports Nvidia's work and uses OpenShell.
  • Proprietary Nvidia Sentry on BlueField-4 DPUs monitors agents from hardware they cannot see.
  • OpenAI points to its own safeguards and the Defense Factory info-sharing effort.

Who signed and who didn't

TechCrunch's Julie Bort reports that Nvidia's consortium to contain rogue AI agents, the Open Agent Safety Platform, counts more than 100 companies, including Anthropic as a supporter. Missing from the public list are OpenAI, Amazon, Google and Apple. An OpenAI spokesperson said the company supports Nvidia's work, and TechCrunch notes OpenAI already uses OpenShell, the platform's sandboxing technology.

Hugging Face CEO Clem Delangue, whose company Nvidia agreed to acquire for $12.9 billion, said the stack might have caught OpenAI's agents before they breached Hugging Face. The platform launch is covered in Nvidia launches new platform for reining in rogue agents.

What the platform includes

The platform distributes largely open-source agent-security technology, including OpenShell sandboxes, which can be adapted for chips from other vendors; Arm and Intel have signed on. Full enforcement, however, uses proprietary Nvidia Sentry software running on BlueField-4 data processing units, which monitor agents from hardware the agents cannot observe. Nvidia describes the approach in its reference for in-silicon agent monitoring.

That mix of open software and proprietary hardware enforcement may help explain hesitancy among large companies with their own chips or cloud platforms. Joining could mean endorsing a design whose strongest protections depend on Nvidia hardware.

OpenAI's alternative approach

OpenAI points to its own research safeguards and disclosures, its participation in Defense Factory, an information-sharing effort with Anthropic, AWS and Google, and commercial cybersecurity offerings such as Daybreak. It has faced intense scrutiny after a string of agent incidents, and coverage has described one company at the center of a wave of rogue AI attacks.

The absence matters because agent containment depends partly on shared standards. If the largest model developers and cloud providers use different frameworks, interoperability and coordinated response could suffer. Whether OpenAI or others join formally, how widely OpenShell is adopted on non-Nvidia hardware and whether Defense Factory publishes results are the signals to watch. More in our agents coverage and cybersecurity coverage.

Who feels it

Agent developers
OpenShell offers open sandboxing, with full enforcement tied to Nvidia hardware.
Cloud providers
Joining may mean endorsing an Nvidia-centered design.
Security teams
Fragmented standards complicate coordinated response.

What to watch

  1. Formal participation by OpenAI, Google, Amazon or Apple.
  2. OpenShell adoption on non-Nvidia chips.
  3. Published results from Defense Factory.

FAQ

Is OpenAI part of Nvidia's platform?

Not publicly, though it says it supports the work and uses OpenShell.

Who else is missing?

Amazon, Google and Apple.

Is Anthropic involved?

Yes, as a supporter.

What is Nvidia Sentry?

Proprietary software on BlueField-4 DPUs that monitors agents from hardware.

Read the original

Continue at the source.

TechCrunch

Companies: OpenAI, NVIDIA

More on OpenAI

All OpenAI coverage →