OpenAI CEO Urges Global AI Regulation
OpenAI CEO Sam Altman is calling for urgent global regulation for superintelligence, comparing the need to that of the International Atomic Energy Agency (IAEA). The call comes despite reports of OpenAI's prior lobbying against stricter AI rules in the U.S. and E.U. Altman suggested that such advanced AI could emerge within years, necessitating a proactive international framework.
- In lobbying related to the EU's AI Act, OpenAI successfully argued against classifying general-purpose AI systems like GPT-3 as "high-risk," a designation that would have imposed stricter legal requirements for transparency and human oversight. This aligned with the positions of Google and Microsoft, shifting the regulatory burden to companies that apply AI to high-risk use cases rather than the developers of the foundational models. - The proposed "IAEA for AI" would be an international authority to inspect advanced AI systems, conduct audits, and place restrictions on deployment to manage the development of "superintelligence." This call for regulation of future, more powerful AI contrasts with the company's efforts to lessen the regulatory burden on its current systems. - OpenAI's federal lobbying spending in the U.S. reached $2.99 million in 2025, focusing on shaping legislation around copyright liability, AI safety standards, and intellectual property. The company has a four-person in-house lobbying team with deep roots in Congress and also retains outside counsel from firms like DLA Piper and Akin Gump. - In the U.S., OpenAI advocates for a unified federal regulatory approach, arguing that a "patchwork" of 50 different state-level regulations would stifle innovation and hinder competitiveness with China. The company has specifically pushed for federal preemption, which would override state laws. - Sam Altman's recent call for urgent regulation at the AI Impact Summit in New Delhi included warnings about specific threats, such as the potential for open-source AI models to help create new biological pathogens. - The EU AI Act, which will largely become effective in August 2026, establishes a risk-based framework. While most obligations for "high-risk" systems are two years away, bans on "prohibited practices," such as certain uses of biometric identification and emotion recognition, will take effect in February 2025. - At the New Delhi summit, UN Secretary-General António Guterres echoed the need for regulation but also cautioned that the future of AI "cannot be decided by a handful of countries — or left to the whims of a few billionaires," calling for a $3 billion fund to ensure open access for developing nations. - OpenAI's own timeline predicts AI systems will be capable of making "significant discoveries" by 2028, underscoring the company's view on the rapid advancement toward more powerful AI. This forecast is a key driver behind the call for a proactive global framework to govern future "superintelligent" systems.
