Ω NEWS BRIEF
SCIENCE DESK — DISCOVERY REPORT
Tuesday, September 8, 2026
OPENING SIGNAL
A machine did not invent a new proof of Fermat’s Last Theorem. It did something rarer and more operational: it turned an existing human proof into thirteen million lines of Lean that a kernel can check, in eleven days, with no extra axioms. [T] Kevin Buzzard compiled the artefact and said the statement matches Mathlib. That is not “AI discovered math.” It is AI becoming a verification engine for the library of recorded science. Hold that next to a pig kidney that kept a man off dialysis for 271 days and then handed him to a human graft. [T] The week’s signal is not magic. It is instrumentation: tools that check proofs, and organs that buy time.
1. THE BIG DISCOVERIES
[T] Claude formalizes Fermat’s Last Theorem in Lean
WHAT HAPPENED: Anthropic reported on 4 September that a multi-agent Claude run produced the first end-to-end computer-checked formalization of FLT in Lean. About 13 million lines, roughly 29,500 intermediate theorems used in the finished argument, Lean’s three standard axioms only. Nature and New Scientist covered it 5–7 September. This is autoformalization of the Darmon–Diamond–Taylor exposition of the Wiles–Taylor path, not a new number-theoretic idea.
WHY IT MATTERS: Formalization has been a years-long human bottleneck. A compiled, kernel-checked artefact changes what “we know this proof is complete” can mean.
WHAT IT COULD BECOME: A template for grinding the modern literature into machine-checkable form. [I]
WHAT IT MEANS FOR YOU: If mathematics can be audited at this scale, so can other dense technical claims. Trust shifts from reputation toward replayable check.
[T] Pig kidney as a 271-day bridge to a human transplant
WHAT HAPPENED: The Lancet case from Mass General Brigham: Tim Andrews received a gene-edited Yucatan miniature pig kidney on 25 January 2025. It functioned 271 days, then was removed; after 82 days of dialysis he received a human kidney in January 2026. First documented xenograft-to-allograft sequence of this kind. A later eGenesis expanded-access patient was reported past that duration; treat that as a separate, still-evolving case. [T]/[U]
WHY IT MATTERS: Dialysis is not a holding pattern. It grinds the body. A working xenograft that lasts months changes the waiting-list physics.
WHAT IT COULD BECOME: Temporary bridges first; durable replacement only if rejection, infection, and zoonosis stay controlled. [H]
WHAT IT MEANS FOR YOU: Organ scarcity is an engineering and immunology problem, not only a donation slogan.
[T] In-plane anomalous Hall effect
WHAT HAPPENED: Carnegie Mellon, Nature Materials (reported 3 September): a Hall-type voltage appears when magnetization and field sit in the plane of the film, not only when the field is perpendicular. Century-old classroom assumption: wrong as a universal rule.
WHY IT MATTERS: Geometry of magnetic readout just opened. Vector magnetometry in one planar device becomes a design path. [I]
WHAT IT COULD BECOME: Simpler sensors in electronics, transport, and imaging if the effect is large and clean enough to engineer. [H]
WHAT IT MEANS FOR YOU: Most of the devices around you already use Hall physics. Extra axes mean extra information from the same chip family.
[T] Soil enzyme that cuts bioplastic and β-lactam antibiotics
WHAT HAPPENED: ISME Journal: forest-soil microbes on long-chain aliphatic polyester (LCAP) yielded family-VIII esterase LCPH1. Heterologous expression confirmed dual activity: polyester depolymerization and hydrolysis of penicillin and ampicillin. Structure sits near type-C β-lactamases; wide “pac-man” active site.
WHY IT MATTERS: The plastisphere already looks rich in resistance genes. An enzyme that eats both substrate classes is a mechanistic link, not a slogan. [I]
WHAT IT COULD BECOME: Useful recycling catalyst — or another route by which plastic debris selects for antibiotic-defanging chemistry. Both can be true. [H]
WHAT IT MEANS FOR YOU: “Biodegradable” is not a moral category. It is a reaction network in dirt.
[U] One WIMP-like scatter in LZ is not a discovery
WHAT HAPPENED: LUX-ZEPLIN reported a single xenon nuclear-recoil event whose timing and light look like the WIMP template. Team, speaking at a Japan meeting and in a PRL submission, said explicitly: one event, not a claim of dark matter. [T] for the statement; [U] for the particle identity.
WHY IT MATTERS: The search is now in the regime where a single pretty event can leak into headlines.
WHAT IT COULD BECOME: Either the start of a statistical pile, or a background fluke. Only more exposure decides. [H]
WHAT IT MEANS FOR YOU: Dark matter remains the dominant mass budget of galaxies. One flash does not close the ledger.
[T] JWST rules out 0.1-Earth-radius moons around LP 890-9c
WHAT HAPPENED: arXiv:2609.05301, David Kipping, 4 September. Twelve JWST transits of a rocky temperate planet. No exomoon. 95% exclusion of 0.1 R⊕ moons across the Hill sphere — Europa/Rhea/Umbriel scale. Tides on a 0.04 au orbit make large moons unlikely to last anyway.
WHY IT MATTERS: Sensitivity, not a moon. JWST can now hunt at Solar-System moon sizes if the geometry cooperates.
[T] Roman launched; BepiColombo dropped its cruise stage
Roman Space Telescope left Kennedy on a Falcon Heavy, 30 August 2026. Wide-field IR survey for dark energy, dark matter, and microlensing planets. [T] BepiColombo’s Mercury Transfer Module separated 3 September 2026; ESA confirmed signal and healthy MPO power. Orbit insertion targeted 21 November; Mio and MPO split in December. [T]
2. SPACE / COSMOS
Roman is the week’s hardware change: a Hubble-class sharp eye with a field at least a hundred times wider, built to count rather than portrait. [T] BepiColombo is now in arrival choreography after a year slip from electric-propulsion trouble. Mercury still has no two-orbiter science constellation; that starts after November. [T]
Hubble imaged the N44 superbubble in the Large Magellanic Cloud — a known star-forming complex, newly framed, not a new class of object. [T] Saturn’s south pole now has a reported decagonal wave to sit beside the famous north hexagon; treat morphology as observation, mechanism as open. [T]/[H]
A globular-cluster stellar stream in ultra-diffuse galaxy UGC9050-Dw1 (Niels Bohr Institute, mid-August) is the first such stream seen outside the Milky Way and a new dark-matter weighing tool. [T] for the stream; [I] for the mass model.
Exomoon null result above is the cleanest planetary-science limit of the week. Little-red-dot and early-galaxy debates continue from Webb’s four-year census; those are multi-year tensions, not a 24-hour flip. [I]
3. PHYSICS / QUANTUM
The CMU in-plane Hall result is experimental condensed-matter, not a thought experiment. [T] It does not repeal the ordinary Hall effect. It adds a geometry.
LZ’s single candidate remains below discovery threshold. [U] Solar-neutrino work at XENONnT at record-low energies (Weizmann/XENONnT, late August) is a separate, cleaner measurement of a known flux, not dark matter. [T]
Laser-sail papers this month note that at roughly 0.75 c, scattered photons can add drag — a relativity bookkeeping problem for interstellar concepts, not a flying prototype. [I]
Do not promote “hairy black hole listening” methods or ringdown templates from secondary write-ups without the paper in hand. Limited data on that item this cycle. [U]
4. AI + SCIENCE
FLT autoformalization is the scientific-instrument story, not a chatbot product launch. Agents wrote Lean; Lean checked it; a human specialist compiled it. [T] Buzzard’s own caveat stands: the mathematics of FLT was already believed; the artefact “faithfully follows the early literature and adds nothing” to the number theory. [T] What changed is throughput and auditability.
A separate Nature note this week: mutating every DNA letter of a viral genome produced effects that leading AI models failed to predict. [T] Sequence models are not oracles of phenotype.
PNAS Nexus editorial (8 September): “augmentation not abdication” for AI in research — norms, not a result. [T] as opinion piece.
OpenAI GPT-6 launch on 3 September is product news. It is not, by itself, a scientific demonstration. Parked.
5. ENGINEERING / MATERIALS / ENERGY
Virginia Tech reports degradable polymers that combine strength, toughness, flexibility, and oxygen barrier in one architecture — properties usually traded against each other. [T] as institutional claim; independent scale-up not shown here. [U]
LCPH1 is an enzyme tool and an ecological warning in the same molecule. Engineering use requires stability, yield, and a polymer it actually meets in waste streams. LCAP is not PET. [I]
In-plane Hall geometry is a materials-device story as much as a physics story: planar vector sensors if signal-to-noise cooperates. [I]
Roman and BepiColombo are flight systems under test. Roman’s science starts after commissioning. BepiColombo still has insertion, orbiter split, and 2027 science orbits ahead. [T]
6. EARTH / LIFE / HUMAN ORIGINS
Xenotransplant case is the medical-biology headline. Gene edits removed antigenic sugars and added human genes; late vascular injury still ended the graft. [T] That failure mode is now a design target.
NEJM (2 September): RAS inhibitor daraxonrasib, already FDA-approved for metastatic pancreatic cancer, shows activity signals in lung cancer. Promise, not a new standard of care from one report. [T]/[I]
Danish nationwide cohort: early statin after type 2 diabetes diagnosis associated with about 15% lower relative dementia risk in 132,000 people. Observational. Confounding is not dead. [I]
Gut-microbiome transmissibility work this week argues some bacterial populations move between people and continents faster than older models assumed. Mechanism and clinical weight remain open. [I]/[U]
115-million-year dinosaur material from a Bayan Gobi drill core extends the local record; it is a locality result, not a rewrite of dinosaur phylogeny. [T]
Africa’s giant herbivores declining millions of years before human hunting is a paleoecology story about climate and vegetation, not a single new bone. [I]
Nature also flags child deaths in Chinese gene-therapy trials and UK Biobank data-leak aftermath. Those are governance facts with scientific consequences, not discoveries. [T]
7. THE PAPER OF THE DAY
ISME Journal — family-VIII esterase LCPH1: polyester hydrolase and β-lactamase in one protein.
QUESTION: What enzymes do soil microbes use on emerging long-chain aliphatic polyester bioplastics, and do those enzymes touch other human-relevant substrates?
METHOD: Year-plus burial of LCAP strips in forest soil; imaging of pits; community sequencing; identification of a family-VIII esterase enriched on the plastic; structure comparison to type-C β-lactamases; heterologous expression; in vitro assays on polymer and on penicillin/ampicillin.
RESULT: Dual activity confirmed. Wide active site. Predicted membrane-associated lipoprotein. Plastic debris communities already known to carry resistance genes now have a concrete catalytic overlap. [T]
LIMITATIONS: One enzyme family, one polymer class, laboratory hydrolysis is not field-scale recycling. Antibiotic cleavage in a tube is not proven clinical resistance transfer. [U] for population-level risk.
WHY IT MATTERS: It ties two public problems — persistent polymers and antibiotic failure — to a single active site. That is a mechanism, not a metaphor.
8. WHY THIS MATTERS TO YOU
If you will ever need a kidney, the Andrews case is the first hard proof that a pig organ can buy nine dialysis-free months and then get out of the way for a human graft. [T] If you work with proofs, models, or regulations, the Lean FLT artefact is a preview of machine-checked technical literature. [T] If you take antibiotics or throw “compostable” plastic in soil, LCPH1 says those chemistries already share enzymes in the dirt. [I] Roman will not change your week. It will change the catalogs astronomers hand the next decade. [I]
9. REALITY CHECK
WHAT WE KNOW: FLT is now machine-checked along a standard path. [T] A gene-edited pig kidney functioned 271 days in a living recipient and was followed by a successful human transplant. [T] An in-plane Hall response exists in the reported films. [T] LCPH1 hydrolyzes LCAP and β-lactams in vitro. [T] Roman is in space. BepiColombo dropped MTM. [T] JWST can exclude 0.1 R⊕ moons around LP 890-9c. [T]
WHAT WE THINK: Autoformalization will eat large parts of existing math faster than human projects. [I] Xenografts will be bridges before they are destinations. [I] Dual-use plastic/antibiotic enzymes are common enough to matter for resistance ecology. [H]
WHAT WE DON'T KNOW YET: Whether LZ’s event is a WIMP. [U] Whether daraxonrasib becomes a multi-tumor standard. [U] Whether in-plane Hall is industrially large. [U] Whether LCPH1-type activity accelerates resistance in real waste and clinics. [U] Whether Roman’s first-year catalogs will shift dark-energy constraints or mostly refine them. [U]
HYPE WATCH: “AI proved Fermat” is false. Humans proved it; a model formalized a known route. “Dark matter found” is false. One recoil is a candidate. “Pig kidneys solved the organ crisis” is false. One documented bridge, rejection still won on the clock.
CLOSING SIGNAL
Remember this: verification scaled this week, not miracles. A theorem became replayable code, and a foreign organ became nine months of human time. Everything else on the desk is either a new handle on old physics or a single event still waiting for a second.