Get the app
Longevity2025

Can AI rejuvenate our cells? What was announced, and what was not

In August 2025, OpenAI and Retro Biosciences announced that an AI model helped redesign the Yamanaka factors, the proteins that turn an adult cell back into a stem cell. In cultured cells, the new versions beat the originals, with “greater than a 50-fold” higher expression of stem cell markers. Here is where the idea comes from, what the numbers really say, and why a treatment is still far away.

OpenAI (billet de recherche), 2025Lire en français

Reprogramming a cell: where does the idea come from?

For a long time, a specialized cell (skin, gut, muscle) was thought unable to go back. In 1962, John Gurdon showed that the nucleus of a tadpole intestinal cell, placed in a frog egg, could produce a normal tadpole: the adult cell had kept all the information it needed.

In 2006, Shinya Yamanaka and Kazutoshi Takahashi went further: by introducing only four factors (Oct3/4, Sox2, c-Myc and Klf4) into mouse cells, they obtained induced pluripotent stem cells, called iPS cells. Gurdon and Yamanaka shared the 2012 Nobel Prize in Physiology or Medicine for the discovery that mature cells can be reprogrammed to become pluripotent.

Why can't it be used as is?

According to OpenAI, the original process has poor efficiency: typically less than 0.1% of cells convert, and it can take three weeks or more. Efficiency drops further with cells from aged or diseased donors, which is a problem for anything related to aging.

There is also a safety question. In the 2006 paper, iPS cells transplanted under the skin of mice formed tumors containing a variety of tissues. That is the test that proves pluripotency, but it shows why these cells cannot be used without precautions.

What did OpenAI and Retro Biosciences announce?

On August 22, 2025, OpenAI published a post describing GPT-4b micro, a miniature version of GPT-4o specialized for proteins. With the team at Retro, a longevity biotech company, it was used to propose new versions of two of the four factors, SOX2 and KLF4, called RetroSOX and RetroKLF.

According to the post, over 30% of the model's SOX2 suggestions outperformed the natural protein in a first screen, versus hit rates below 10% in traditional screens. For KLF4, 14 model-generated variants beat the best cocktails of the earlier screen. Combined, they gave the largest effect: late stem cell markers appearing several days sooner than with the original factors.

Key figures

  • over 50-foldhigher expression of reprogramming markers with the AI-designed proteins than with the originals, in the lab
  • under 0.1%of cells convert with the original process, over three weeks or more
  • over 30%of cells from donors over 50 expressing stem cell markers within 7 days
  • 14 variantsof KLF4 better than the best cocktails of the earlier screen

What science may deliver someday, Mima helps you track today

Aging research moves slowly, but your energy, your sleep and how you feel today are things you can observe right now. With Mima, you record a voice note of a few seconds each day, and you get a clear summary of what changed, week after week.

You can try it for free. Before an appointment, Mima also helps you prepare questions for your doctor, without replacing their advice.

Try Mima for free

What do “50-fold” and “30% in 7 days” really mean?

The “greater than 50-fold” figure refers to the expression of reprogramming markers in the lab, compared with the original proteins. It is not a measure of rejuvenating an organism, nor 50 times more converted cells: it is a lab indicator.

The “30% in 7 days” figure comes from another test, run with messenger RNA on cells from three donors over 50: within 7 days, more than 30% of cells began expressing stem cell markers. So the cells were not “repaired”: they started to look like stem cells. According to the post, over 85% of the cells then activated their own stem cell markers.

And what about rejuvenation?

OpenAI also measured DNA breaks, one of the hallmarks of cell aging. After a stress that damages DNA, cells treated with the RetroSOX and RetroKLF variants showed fewer breaks than cells treated with the original factors, with modest statistical differences (p between 0.01 and 0.04 depending on the comparison).

The authors conclude that the variants offer a “potential path” toward improved cell rejuvenation. That is a research hypothesis, not a result demonstrated in a living being.

What these results do not say

The post is published by OpenAI, which notes that Sam Altman is an investor in Retro Biosciences. It does not cite a peer-reviewed journal publication, and the experiments were done in cultured cells.

Nothing indicates that a treatment for people is close, or that a person can be “rejuvenated”. Going from a lab experiment to a treatment takes safety studies (notably on tumor risk), then human trials, which take years. If stronger results are published, this page will be updated.

Frequently asked questions

Can AI rejuvenate our cells?

Not in humans at this stage. OpenAI and Retro Biosciences report that AI-designed proteins reprogram cultured skin cells better and reduce some DNA breaks. These are lab results, not published in a peer-reviewed journal.

What are the Yamanaka factors?

Four proteins (Oct3/4, Sox2, Klf4 and c-Myc) that, introduced into an adult cell, turn it back into a pluripotent stem cell. Shinya Yamanaka identified them in 2006 and received the Nobel Prize in 2012 with John Gurdon.

What does “50 times more efficient” mean?

According to OpenAI, the redesigned proteins gave more than 50-fold higher expression of reprogramming markers than the original proteins, in the lab. It is not an effect measured in an organism.

Are 30% of cells repaired in 7 days?

No, that is not what the source says. In a test on cells from three donors over 50, more than 30% of cells expressed stem cell markers after 7 days. That is reprogramming, not repair.

What are the risks?

iPS cells can form tumors in the standard mouse test. Safety has to be shown before any use in humans, which has not happened.

Is a treatment available?

No. The results are in cultured cells. The source mentions no human trial.

The original study

OpenAI et Retro Biosciences “Accelerating life sciences research” OpenAI (billet de recherche), August 22, 2025.

Read the OpenAI post

This article summarizes a scientific study and is not medical advice.