Why OpenAI did not publicly join Nvidia’s AI agent safety group

In short: OpenAI did not publicly sign on to Nvidia’s new AI agent safety effort, but it is still working with Nvidia behind the scenes.

What happened

Nvidia announced a new group of more than 100 companies focused on stopping “rogue AI agents.” AI agents are programs that can take actions on their own, like clicking around websites or running tasks for a user.

The effort is called the Open Agent Safety Platform. Nvidia says it is meant to reduce incidents where agents do things they should not do, such as bypassing rules or coordinating harmful actions.

OpenAI did not appear as a public supporter. TechCrunch reported that OpenAI, along with Amazon, Google, and Apple, has not joined the public list. An OpenAI spokesperson told TechCrunch the company supports Nvidia’s work.

TechCrunch also reported that OpenAI is working with Nvidia on agent security, including a key piece of software called OpenShell. OpenShell is open source software that creates a “sandbox” (a locked room for code) designed to keep agents from escaping and causing trouble.

One reason some companies may hesitate to publicly commit is that Nvidia’s full system includes a proprietary hardware part that only runs on Nvidia equipment. Nvidia says its monitoring feature, called Sentry, runs on special Nvidia processors and can watch agent behavior and shut agents down.

Why it matters

As more apps add AI agents, mistakes and misuse can affect real people, like when software hits the wrong websites or breaks rules. This story shows that safety work is becoming both a shared goal and a business decision, since some solutions may tie customers more closely to one company’s hardware.

Source: TechCrunch AI