Since this story's prior coverage on June 5, the OpenAI narrative has taken two separate but connected turns — one about government control, one about whether humans are still running the show at all.

OpenAI Agrees to Pre-Release Federal Review

President Trump signed an executive order requiring AI companies to submit their frontier models for government benchmarking 30 days before public release. The order is voluntary. OpenAI said yes anyway.

George Osborne — OpenAI's Head of Countries and the former U.K. Finance Minister under David Cameron — told CNBC's Arjun Kharpal at SXSW London that OpenAI will comply. His quote: "It's quite right that democratic governments have a big role to play in how this technology is used and deployed."

Osborne went further. He said OpenAI has proactively suggested oversight mechanisms to governments — not just in the U.S. — rather than waiting to be regulated. "We don't wait to be asked," he said.

The benchmarking process under Trump's order is specifically designed to assess "advanced cyber capabilities" and determine when a model crosses the threshold to be designated a 'covered frontier model.' That designation presumably triggers additional scrutiny.

OpenAI is welcoming federal eyes on its most powerful products before those products go live. Whether that's savvy PR, genuine safety commitment, or both — it's a departure from the standard Silicon Valley posture of 'regulate us later.'

The Bigger Story: AI Is Now Building AI

SoftBank CEO Masayoshi Son told CNBC on Monday that he spoke directly with OpenAI CEO Sam Altman and engineers at the firm. What they told him: an AI model is currently designing OpenAI's next model.

"Engineers will no longer be smart enough to design the next model," Son said. "So once that happens, the model generates the next model — and it's going to be exponentially smarter than all of us. That's a superintelligence."

OpenAI declined to comment on unreleased models. But the company has already acknowledged a version of this. OpenAI has previously stated that it used early versions of its models to assist in debugging training, managing deployment, and evaluating test results — a recursive process in which AI contributes to the creation of successor models.

So this isn't entirely speculation. The recursive loop has already started.

Son's Timeline: Two Years to Superintelligence

Son's forecasts on artificial superintelligence — AI that is, by his definition, 10,000 times smarter than humans — have been accelerating dramatically. In 2024, he predicted ASI in 10 years. He told CNBC on Monday he was "trying to be conservative" at the time. His real estimate was four years. Now he says two years.

The man who runs one of OpenAI's largest outside shareholders is saying human-level-and-beyond AI arrives by 2028 — and he's basing that partly on what Altman's own engineers told him.

Meanwhile, Anthropic — the company that builds Claude — made a public statement this week with the opposite message: AI development should slow down to deal with the implications of how fast things are moving. Anthropic didn't name OpenAI directly, but the contrast is impossible to miss.

The Gaps in Coverage

Coverage has largely treated Trump's pre-release review requirement and reports of AI-designed AI as separate developments. They intersect in an important way: A company just agreed to let the federal government preview its models 30 days out — while credible sources close to that company say the next model is being built by a machine, not a human team. How exactly does a government benchmarking process evaluate a model designed by another model, on a 30-day preview window, when the underlying capability curve is accelerating faster than the review framework was designed to handle?

The Trump executive order was built for today's AI. If Son and Altman's engineers are right, it may already be obsolete.

Anthropic's warning has also received less attention than it deserves. The company employs some of the most credible AI safety researchers in the world, and it's calling for a slowdown. That deserves equal weight alongside Son's bullish timeline.

What This Means for Regular People

If AI is designing AI, the pace of capability improvement is no longer constrained by how fast human engineers can work.

The federal oversight mechanism Trump signed this week is a start. But 30-day pre-release benchmarking wasn't designed for a world where the thing being reviewed was built by the previous version of itself. Washington needs to catch up — fast. And OpenAI agreeing to play ball voluntarily doesn't change the fundamental question: who actually controls what comes next?