OpenAI skips Nvidia agent safety consortium

OpenAI has declined to join Nvidia’s consortium of over 100 companies launching the Open Agent Safety Platform, despite being central to the initiative's motivation following recent rogue agent incidents. While competitors like Anthropic and Amazon have signed on, OpenAI is pursuing an independent path through its own Defense Factory cybersecurity consortium and proprietary tools like the Daybreak model. Nvidia’s platform relies heavily on proprietary BlueField-4 hardware for monitoring, a dependency that may explain some hesitation, but OpenAI’s choice highlights a broader industry trend of treating AI safety as a distinct product category. This divergence means enterprises will likely face a fragmented landscape of competing security standards from major AI providers.