Today's issue is supported by Chartsy
Chartsy turns your Stripe or Paddle data into answers. Ask "show MRR by plan" or "why did churn spike" in plain English and get the chart instantly - no spreadsheets, no pivot tables. Built for founders who'd rather read their numbers than wrestle them.
OpenAI has set itself a target. By 2028, it wants to build an AI that can do the job of an AI researcher. That's the person who figures out how to build the next smarter AI.
In plain English
Today, humans build AI. Engineers design it, train it, test it and fix it. The plan is to hand that job to the AI itself. Then each version helps build a better next version, which helps build a better one after that. Researchers call this loop "recursive self-improvement." Think of a factory that starts building better factories.
That sounds like science fiction. It isn't, because it has already started.
2028
OpenAI's target for a fully automated AI researcher
1% → 26%
Share of Anthropic's AI research done by AI with light human oversight, in five months
20+
Senior researchers who signed the warning, Hinton and Bengio among them
~1 year
How long Hinton says Congress has to act
It's already doing a quarter of the work
On September 28, more than 20 researchers published a paper called "What if automating AI R&D triggers an intelligence explosion?" The authors include Geoffrey Hinton and Yoshua Bengio (two of the three "godfathers of AI"), OpenAI's chief scientist, Microsoft's chief scientific officer and an Anthropic co-founder. These are the people building the systems, not outside critics.
Their key evidence is one number. As of August, Anthropic reported that its AI systems were completing 26% of the company's internal AI research work with only high-level human supervision. Five months earlier, that figure was 1%. The paper also says AI now writes most of the code inside the companies that build it, and that OpenAI reports its systems routinely finishing research tasks that would take its staff days.
One quarter is still a minority. The number to watch is how fast it moved.
Why scientists are alarmed
If AI can do research, one lab could copy that AI thousands of times and run it day and night. The authors say that could squeeze years of progress into months. Their concern isn't just the technology. It's the timing. If progress speeds up that much, governments and companies might not have time to react, and the window to act could close.
Their asks are modest and specific: regular public reporting on how much AI research is being automated, independent auditors placed inside AI companies, and ways to pause projects if something goes wrong.
Geoffrey Hinton | Told US senators on Sept 16 that Congress has about a year to act, and not much longer |
Yoshua Bengio | Asked the UN Security Council on Sept 23 to license top AI systems the way nuclear power and aviation are licensed |
OpenAI | Targets a fully automated AI researcher by 2028 |
Anthropic's CEO | Told the same UN meeting the company will slow down as much as necessary |
President Trump | Rejected international plans to control AI in his UN speech |
Trump and Xi | Agreed an AI incident hotline on Sept 25. US officials said chip export controls were not on the agenda |
Keep this honest
2028 is a target, not a prediction. It's what OpenAI is aiming for. It doesn't mean it will happen.
The 26% needs context. It's Anthropic's own number, and "high-level supervision" still means humans are involved.
The authors don't say an explosion is certain. They list real brakes: limited computing power, hardware supply chains, regulation, and diminishing returns. Hinton's one-year deadline is his personal estimate, and the Trump administration has pushed back on the warnings.
🔮 The Bottom Line
OpenAI has put a date on handing AI research to AI. The world's two AI powers responded with a hotline for emergencies, not a brake.
A target isn't a forecast. But this one is written down, and the people warning about it work at the companies chasing it.
That's this week. Hit reply with your read: is 2028 realistic, or hype? We'll share where readers land next week. Our video breakdowns hit Instagram first at @unseen_ainews.
- hiPreneurs
📧 Forward this to 3 entrepreneur friends who need to see this opportunity
