Tech • AI • Robotics
AI Just Did the Impossible: Reversed Human Aging
9.4/10TL;DR
An AI-designed drug for idiopathic pulmonary fibrosis unexpectedly made patients’ blood appear 3 to 6 biological years younger across six independent aging clocks, while separate advances in AI agents, UN warnings, and Mistral’s record funding underscored the sector’s widening impact.
KEY POINTS
Drug trial produced an unexpected aging signal
Insilico Medicine, based in Hong Kong, developed Rentosertib to treat idiopathic pulmonary fibrosis, a disease that scars lung tissue. In a trial involving 42 patients, the drug improved lung measures, but a later analysis of routine blood samples found something researchers were not targeting: multiple AI-based aging models judged treated patients biologically younger than those on placebo.
Six independent clocks agreed
The blood data was tested with six aging clocks built by separate teams, including groups at Harvard, Oxford, Beijing, and Insilico. All six pointed in the same direction, with the effect peaking at week four at roughly 3 to 4 years younger, and one model estimating as much as 6 years. The result does not mean patients physically became younger, but that their blood protein patterns shifted toward profiles these systems associate with younger biology.
Why researchers are taking it seriously
Interest centers less on the magnitude than on the consistency. Nobel laureate Michael Levitt highlighted the agreement among clocks trained on different data and features, arguing that six independent systems all being wrong in the same direction would be harder to dismiss than a single model result. The findings were reported in Nature Biotechnology.
Not simply a byproduct of healthier lungs
A key objection is that patients may have looked younger on blood tests simply because their lung disease improved. But the dose with the strongest lung benefit was 60 milligrams once daily, while the largest drop in predicted biological age came at 30 milligrams twice daily. Because the best dose for lungs and the best dose for age signals did not match, the data suggests at least part of the effect may be independent of lung improvement.
Protein changes matched known aging patterns
Researchers also compared treated patients’ protein shifts with more than 55,000 samples from the UK Biobank. The drug appeared to reverse protein changes that normally move in a consistent direction with age, rather than causing random movement across biomarkers. Even so, the company describes the result as potential rather than proof, and no study has yet shown that such clock changes translate into longer life or broader anti-aging benefits.
Experts urged caution and broader testing
Outside specialists said the findings are promising but preliminary. Eric Topol described the drug as encouraging while stressing the lack of a definitive trial. Vadim Gladyshev of Harvard Medical School noted the small sample and limits of biological clocks, but also called it the first study to show clearly that predicted biological age can be reduced. The next obvious test is whether the effect appears in healthy people rather than patients with a specific disease.
AI compressed the drug-discovery pipeline
The case is also a demonstration of AI-assisted drug design. Insilico uses one system to scan scientific and biomedical data for disease-related targets and a second to generate molecules tailored to those targets. That process led to a protein called TNIK, linked to both pulmonary fibrosis and aging, and produced a candidate in about 18 months, far faster than conventional timelines. Rentosertib is now in Phase 3 trials, and the company says it has generated at least 28 candidates through the same approach.
AI agent cleared every level of a CAPTCHA game
In a separate development, GPT-6 Astra completed all 48 levels of I Am Not a Robot, a difficult web game built around increasingly absurd CAPTCHA tasks. The model used full computer control rather than direct image inputs, reading screens, moving the cursor, typing, dragging sliders and adapting to shifting rules. The feat suggests AI systems are getting closer to defeating practical tests designed to separate humans from bots.
Benchmarks and security concerns intensified
On OSWorld 2.0, which measures real computer use, Astra scored 72.6% and averaged about 40 minutes per task, versus 65.7% and 75 minutes for the prior generation. OpenAI has said the model is the first to approach a critical line on its cyber-risk scale, while also reporting strong exploit capability under controlled testing. Those gains fed into wider concerns that human-verification tools and current safety measures may not keep pace.
UN warning and Mistral’s record raise
In Geneva, UN human rights chief Volker Türk warned that advanced AI could pose an existential risk to humanity and called for international red lines and stronger safety guarantees. Meanwhile, Mistral raised €3 billion at a valuation of about €24 billion, described as the largest private tech equity round in Europe. The French company said the money will go to frontier research and model capability, while executives framed the raise as part of Europe’s push for AI sovereignty.
CONCLUSION
The unexpected Rentosertib data has not proved that aging can be reversed, but it has produced one of the clearest signals yet that an AI-designed medicine may shift biology in that direction. At the same time, rapid advances in autonomous AI systems and record strategic investment are intensifying pressure for stronger governance.
Explain this