OpenAI Has More Powerful AI Than Its Customers Can Use

In Brief

OpenAI researcher Noam Brown acknowledges that the company's unreleased AI creates an unfair advantage. 

As testing takes longer, the gap between what labs can use and what businesses can buy could grow, raising questions about who gets to compete with the strongest AI.

What Happened

OpenAI has an unreleased, general-purpose model that has produced solutions to multiple unsolved math problems, according to OpenAI researcher Noam Brown. In his interview with Dwarkesh Patel, Brown acknowledged the unfair advantage that creates and said he does not know how to balance the competing concerns around access and safety.

The testing problem is becoming harder. Brown described a future in which AI agents can work effectively for three months. (Agents are systems that can autonomously go through multiple steps to complete assignments.) If new models arrive every two months, a lab cannot evaluate a full-length assignment before the next release cycle.

Patel raised a further possibility: labs could retain their strongest models and use them to develop even better AI. That is simply a scenario, not an announced plan.

SmarterX founder and CEO Paul Roetzer examined the implications for competition and business planning on Episode 241 of The Artificial Intelligence Show.

The Key Numbers

130 - Billions of tokens or chunks of text processed by AI in one of OpenAI's math efforts 

Less than 10% - Credit Brown would assign to multiple agents for that math result

<2  - Interval between leading-model releases in months described by Brown

3 - Length, in months, it takes an AI agent to complete a task in the future, according to Brown's safety-testing example

Stronger Models Solving Bigger Problems

The underlying model matters more than the number of AI agents. Teams of agents can divide a problem and work simultaneously, much as employees split a research project. But Brown cautioned against crediting the math result mainly to that arrangement. His central point was that OpenAI had trained a stronger model. Reproducing the team structure would not give another business the same capability.

Slower releases can leave internal advantages intact. Roetzer extended the discussion to a business scenario: a lab could use private AI to build firms in industries such as consulting while competitors rely on publicly available tools. But release schedules alone cannot answer the competition question.

"I do think this concentration of power resulting from unreleased models only being available to select labs and their partners, including the government, is a way more real thing than the pDoom stuff."

— Paul Roetzer, founder and CEO of SmarterX on Episode 241 of The Artificial Intelligence Show

By pDoom Roetzer means estimates of AI causing catastrophe. He sees unequal access as a more immediate concern, while acknowledging that the right response remains unresolved. "I don't know the answer here," he says.

Safety work can also affect development. "Pacing probably also means way more resources dedicated to safety and alignment internally," Roetzer says. More testing and work to keep agents following human instructions could slow internal progress, too.

SmarterX Take

A business can assess the AI it uses today and still underestimate what is coming. A public product shows what a lab has chosen to release. Leaders should distinguish that experience from researchers' accounts of internal capabilities, which need scrutiny.

"We still have to look ahead, and you in your business have to do the same thing," Roetzer says. The practical response is to revisit your strategy as capabilities change: Which work becomes easier? Who gains access first? Could a technology supplier eventually become a competitor?

What to Watch

Access decisions will shape the business impact. Follow which capabilities reach customers, which remain internal, and what evidence labs provide about their testing. Slower public releases would not, by themselves, establish that development has slowed or that the advantage has disappeared.

Do Businesses Have a Plan for Better AI?

Only 29% of respondents say their organization has an AI roadmap, according to the 2026 State of AI for Business Report. That planning gap matters when today's available tools may offer an incomplete picture of tomorrow's competition. A roadmap gives leaders somewhere to translate new capabilities into priorities and revisit assumptions as the evidence changes.

Based on more than 2,100 responses from professionals across roles, functions, and industries, the report examines adoption, readiness, training, and governance. It provides a benchmark for the organizational preparation businesses can control, even when the labs' next release remains uncertain. Read the full report →