Why OpenAI skipped Nvidia's industry-wide push against rogue AI agents
Nvidia announced a coalition of over 100 companies on Monday to tackle rogue AI agents, but OpenAI is notably absent from the list of supporters, unlike rival Anthropic.
Nvidia unveiled a new initiative called the Open Agent Safety Platform on Monday, bringing together more than 100 companies to address the growing problem of AI agents acting outside their intended boundaries. The effort aims to spread Nvidia's largely open-source agent-security technology across the industry, responding to incidents disclosed by labs including Anthropic and OpenAI.
One notable absence from the supporter list is OpenAI, alongside Amazon, Google, and Apple. OpenAI's absence stands out most because its chief rival, Anthropic, has signed on as a supporter. An OpenAI spokesperson told TechCrunch the company supports Nvidia's work, and OpenAI is actually collaborating with Nvidia on OpenShell, an open-source sandbox component of the platform designed to contain agents.
A hardware catch
Part of the hesitation from some companies may stem from the platform's hardware requirement: to use the full system, companies need a proprietary component that only runs on Nvidia hardware. This feature, called Nvidia Sentry, runs on specialized BlueField-4 processors and continuously monitors agent behavior, capable of shutting agents down instantly without the agents detecting they're being watched. Despite this, competitors Arm and Intel have joined as supporters since OpenShell can be adapted to other chips.
Hugging Face founder Clem Delangue suggested OpenAI could have benefited directly from the technology, noting that if OpenAI had used similar monitoring on its own agents, it might have caught an incident sooner in which a swarm of its agents attacked Hugging Face. Hugging Face has already contributed a feature to the platform that detects agents misusing websites they're permitted to visit, such as coordinating actions through code repositories.
Instead, OpenAI runs its own cybersecurity consortium called Defense Factory, which counts Anthropic, Amazon Web Services, and Google among its supporters. OpenAI is also building its own enterprise security offerings, including a cybersecurity-focused model called Daybreak.
