A model from @OpenAI proved that the zeros of the Riemann zeta function sit in a narrower band than mathematicians have known since 1859. A computer checked a coded version of the claim, but people are still checking that the code says what they mean.

A popular blood formula, PhenoAge, often reads cancer patients as biologically older. Testing it in 3,000 surgical patients, a team at @MSKCancerCenter found it rose with tumor stage and did no better than a plain sum of its tests, so it may be reading the cancer, not aging.

A new book, praised in a @bmj_latest column, says no drug has ever slowed human aging. Biologist @jpsenescence agrees on the hype but says aging can still be changed: rapamycin lengthened mouse lives in three labs, and nobody has run that test in people.

Senescent cells pile up with age partly because the immune cells meant to swallow them lose their grip. In a @NatureAging study of old mice, Ana Maria Cuervo's team found these cleaners' own protein recycling fades, and turning it back up restored their appetite.

When may a program act for you after you die? In an essay on @habr_com, Aleksandr Samoylenko argues only with a narrow mandate you gave in life, a log outsiders can audit, a rule to stop and ask human trustees, and living people free to say no.

Pet owners are offered three ways to buy more time: a longevity pill for old dogs, a clone, or freezing. In @techreview, @JessHamzelou argues only the pill keeps the same animal, and its trial in 1,300 dogs has yet to show it adds time.

A new benchmark from @LatchBio hands AI agents real study data and a researcher's question, then grades the biological answer rather than the code. The best agent got just under half of single-step tasks right, and about one in three chained ones.

ATP falls with age in many animals, and one theory ties that to repair. Alaattin Kaya's group at @UVA fed yeast extra ATP and mother cells made 24 daughters instead of 17, even though ordinary yeast have no known way to take ATP in.

Senator @BernieSanders has a bill before the Senate that would pause large AI models and permanently ban ones that help design biological weapons. It never defines that help or exempts medicine, so a protein designer used in aging research might be caught.

To refute a theorem you need one object that meets its conditions and breaks its result. A MathArena agent searched 656 maths preprints for such objects and flagged 39 as seriously broken, publishing each theorem and candidate object for people to check.