In its November 6, 2025 post, OpenAI describes how artificial intelligence (AI) has exceeded expectations. For a long time, people imagined the Turing test (whose purpose is to assess whether a computer’s behavior or conversation can be distinguished from that of a human) as a major milestone, but when systems passed it, the world did not change very much. AI can now converse and solve complex problems, yet most people see only a chatbot or a better search engine.
Today, AI systems outperform the smartest people in demanding intellectual competitions. Although they have weaknesses, these systems appear to be closer to 80% of the way toward an AI researcher than just 20%. The gap between ordinary use of AI and its actual capabilities is enormous.
Rapid Development of AI Capabilities
In software engineering, AI has progressed from tasks that take a human a few seconds to those that take over an hour. OpenAI expects systems capable of handling tasks lasting days or weeks to arrive soon. Systems that could accomplish work that would take humans centuries remain uncharted territory.
The cost per unit of intelligence is falling rapidly—by an estimated factor of 40 per year in recent years. In 2026, AI could make minor discoveries. By 2028 and beyond, OpenAI anticipates systems capable of making significant discoveries, although this is not certain.
OpenAI emphasizes that progress in AI is surprising and that society is adapting to it. Even with rapid improvements, everyday life will remain similar due to inertia. The future will bring better ways of living to more people, although work will change and the economic transition will be difficult. Even the fundamental socioeconomic contract may change.
Positive Impacts of AI on the World
AI will help us understand health, accelerate progress in materials science, drug development, and climate modeling. It will expand access to personalized education for students around the world. Demonstrating these benefits builds a shared vision in which AI improves lives, not just efficiency.
OpenAI is deeply committed to safety, which it sees as a way to maximize positive impacts and minimize negative ones. It considers the risks posed by superintelligent systems potentially catastrophic. Empirical studies of safety and alignment will inform decisions such as slowing development to better examine systems approaching recursive self-improvement. No one should deploy superintelligent systems without robust alignment and control, which requires further technical work.
Recommendations for a Better Future with AI
OpenAI proposes shared standards and insights from leading labs. Labs should agree on safety principles and share safety research, insights into emerging risks, and mechanisms for reducing competitive dynamics. For example, an agreement on standards for evaluating AI control would be useful. This would be similar to building codes or fire safety standards, which have saved many lives.
Another point is public oversight and accountability proportionate to capabilities, supporting positive impacts and mitigating negative ones. There are two schools of thought about AI. One sees AI as an ordinary technology, similar to the printing press or the internet, for which conventional policy tools are effective. The priorities are innovation, protecting the privacy of conversations with AI, and defending against misuse through systems developed in partnership with governments.
For today’s level of AI, regulatory burdens should be minimal so that the technology can spread. The second school sees superintelligence as something new that requires an innovative approach. This includes close cooperation with executive branches and agencies in multiple countries, such as safety institutes, in areas such as mitigating AI’s role in bioterrorism and using AI to detect bioterrorism, as well as addressing the implications of self-improving AI.
Building an AI Resilience Ecosystem
In both scenarios, it is crucial to build an AI resilience ecosystem similar to cybersecurity for the internet—software, encryption protocols, standards, monitoring systems, and emergency response teams. This will not reduce risks to zero, but it will reduce them to a manageable level.
Another recommendation involves ongoing reporting and measurement of AI’s impacts by labs and governments. Understanding specific impacts will help steer the technology in a positive direction. For example, the impact on employment is difficult to predict because AI’s strengths and weaknesses differ from those of humans.
Finally, OpenAI believes in empowering individuals. Adults should be able to use AI on their own terms within the bounds of society. Access to advanced AI should be a basic utility, like electricity or clean water. The goal is to help people achieve their goals.
