OpenAI is not publicly supporting Nvidia's new consortium aimed at curbing rogue AI agents, despite working with Nvidia on related software. The company is pursuing its own AI safety initiatives and developing an enterprise offering, potentially seeking independence from its major investor.
OpenAI's decision to not publicly join Nvidia's AI safety consortium, while still collaborating on underlying technology, suggests a strategic move to maintain its independence and leadership in AI safety development, potentially impacting the broader adoption and direction of AI security standards.
Nvidia announced a new industry-wide consortium of over 100 companies dedicated to addressing rogue AI agents, but OpenAI was notably absent from the list of signatories. While other major tech players like Amazon, Google, and Apple also did not join, OpenAI's omission was particularly conspicuous given that its rival, Anthropic, is a supporter.
Despite not publicly pledging support, an OpenAI spokesperson stated that the company is supportive of Nvidia's efforts and is collaborating on key software components, including OpenShell, an open-source tool designed to sandbox AI agents and prevent them from escaping. Nvidia's initiative, the Open Agent Safety Platform, aims to disseminate its AI agent-security technology throughout the ecosystem in response to incidents disclosed by frontier labs.
Nvidia CEO Jensen Huang described rogue AIs as an engineering problem that can be solved, with the Open Agent Safety Platform representing his commitment to this solution. Hugging Face founder and CEO Clem Delangue, whose company was reportedly attacked by OpenAI's agents, suggested that OpenAI's own technology could have prevented the incident if it had been in use. Hugging Face has contributed a feature to Nvidia's platform to detect and shut down unauthorized agent behavior.
However, the full implementation of Nvidia's platform involves a proprietary hardware component, Nvidia Sentry, which runs on specialized Nvidia processors and enforces agent behavior at a hardware layer. This proprietary aspect means the platform is not entirely open source and is optimized for Nvidia's hardware. Competitors like Arm and Intel have joined the consortium, as the OpenShell software can be adapted for other hardware.
OpenAI's absence from the public pledge may signal a strategic move to maintain independence from Nvidia, its major investor, and to assert its own leadership in AI safety. The company is developing its own safeguards, disclosing its most significant AI incidents, and has established its own information-sharing consortium for AI cybersecurity called Defense Factory, which includes Amazon Web Services and Google. OpenAI is also reportedly building an enterprise offering focused on AI cybersecurity.
Pick the topics you care about. Get only what matters, on your cadence.