OpenAI Predicts AI-Made Discoveries by 2026 as Intelligence Costs Plunge

OpenAI claims the cost per unit of a given level of AI intelligence has fallen roughly 40x per year over the last few years, and it expects systems capable of making "very small discoveries" by 2026, with "more significant discoveries" in 2028 and beyond. The company laid out these predictions alongside a set of policy recommendations in a newly published essay on AI progress and its consequences.

The essay's central argument: the gap between how most people use AI—chatbots and better search—and what current systems can already do is immense. OpenAI writes that systems able to outperform the smartest humans at some of the most challenging intellectual competitions "seem more like 80% of the way to an AI researcher than 20% of the way," even though AI remains "spikey" and faces serious weaknesses.

In software engineering specifically, OpenAI traces a steep capability curve: in just a few years, AI has moved from tasks a person can do in a few seconds to tasks that take a person more than an hour. Systems handling tasks measured in days or weeks are expected soon. "We do not know how to think about systems that can do tasks that would take a person centuries," the company writes.

Why the pace may not feel fast

OpenAI draws an explicit parallel to the Turing test. When the popular conception of that milestone went by, "many of us thought it was a little strange how much daily life just kept going." The essay argues the same dynamic will persist: even with rapid capability gains in the next few years, "day-to-day life will still feel surprisingly constant; the way we live has a lot of inertia even with much better tools."

The economic transition is another matter. OpenAI acknowledges that "work will be different, the economic transition may be very difficult in some ways, and it is even possible that the fundamental socioeconomic contract will have to change." It frames the upside as widely distributed abundance, with concrete benefits in health, materials science, drug development, climate modeling, and personalized education.

Safety first, with a superintelligence caveat

OpenAI describes itself as deeply committed to safety, which it defines as "the practice of enabling AI's positive impacts by mitigating the negative ones." The company treats risks from superintelligent systems as "potentially catastrophic" and argues that empirically studying safety and alignment can inform global decisions—including "whether the whole field should slow development to more carefully study these systems as we get closer to systems capable of recursive self-improvement."

"Obviously, no one should deploy superintelligent systems without being able to robustly align and control them, and this requires more technical work," the essay states.

Two scenarios, two policy playbooks

OpenAI lays out two schools of thought. In the "normal technology" scenario, AI progresses like past revolutions from the printing press to the internet, society adapts, and conventional public policy works. Under this view, OpenAI argues that AI at today's capability levels "should diffuse everywhere" with minimal additional regulatory burden for most developers, open-source models, and deployments—and it "certainly should not have to face a 50-state patchwork" of US regulation.

In the second scenario, superintelligence develops and diffuses at a speed humanity has not seen. Here, OpenAI expects typical regulation will not accomplish much either. The company says coordination with the executive branch and safety institutes of multiple countries will be needed, particularly for mitigating AI applications to bioterrorism and managing the implications of self-improving AI.

What OpenAI wants built

The essay closes with four recommendations:

• Shared safety principles among frontier labs, including shared safety research, learnings about new risks, and mechanisms to reduce race dynamics—possibly including agreed standards around AI control evaluations. OpenAI compares this to how society developed building codes and fire standards, "which have saved countless lives."
• An AI resilience ecosystem modeled on cybersecurity: software, encryption protocols, standards, monitoring systems, and emergency response teams. That ecosystem did not eliminate risk but reduced it to a level society could live with. OpenAI sees "a powerful role for national governments" in promoting industrial policy to encourage this.
• Ongoing reporting and measurement from frontier labs and governments on AI's real-world impact. OpenAI notes that AI's effect on jobs has been hard to anticipate "in part because today's AIs strengths and weaknesses are very different from those of humans."
• Individual empowerment, with adults able to use AI on their own terms within broad social bounds. OpenAI expects access to advanced AI to become "a foundational utility in the coming years—on par with electricity, clean water, or food."

The stakes for the reader are straightforward: if OpenAI's timeline holds, the window for building the safety and governance infrastructure it describes—before systems capable of recursive self-improvement arrive—is measured in a small number of years.