AI & ML
AlphaGenome Atlas charts every possible single-letter change in the human genome
Google DeepMind has published AlphaGenome Atlas, a precomputed catalogue of the predicted molecular effects of all ~9 billion possible single-nucleotide variants across the human genome, produced by running its AlphaGenome sequence model exhaustively rather than per-query. Nature's news team covered the release independently of DeepMind's own announcement, describing it as a reference resource geneticists can query for variants of unknown significance instead of running the model themselves. The concrete deliverable is the exhaustive precomputed atlas; the predictions remain model output rather than measured effects, so its practical value depends on how well AlphaGenome's accuracy holds up in the non-coding regions where interpretation is hardest.
Nature — DeepMind's new genome 'atlas' charts effects of all 9 billion human gene mutations · Google DeepMind — AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome
Quantum Computing
IonQ announces Superion 256 as the base platform for future systems
IonQ has announced Superion 256, described as its sixth-generation quantum computing platform and the architecture all its future compute products are expected to build on. The announcement is a press release with a product name and a generational claim but no published qubit fidelities, algorithmic-qubit counts, or benchmark results, and the industrial design has not yet been shown. Until specifications appear, the substantive content is the roadmap consolidation rather than any demonstrated capability.
The Quantum Insider — IonQ Launches Superion 256 Quantum Computing Platform
SkyWater spins up a merchant foundry division for quantum chips
SkyWater Technology has created SkyWater Quantum Solutions, a dedicated merchant foundry division for quantum computing, networking and sensing, offering two standardized process platforms: SC250 for superconducting circuits and SP90 for quantum photonic routing. It has signed a joint development agreement with Qolab to fabricate that company's Quantum System-in-Package hardware. Named process platforms and a signed customer put this above the usual foundry-partnership announcement, though no yields, device performance or volumes have been disclosed.
Quantum Computing Report — SkyWater Launches Dedicated Quantum Solutions Merchant Foundry Division
Post-Quantum Crypto
IonQ puts a number on breaking 256-bit ECC: 19,397 qubits, 25.7 days
IonQ has released an end-to-end fault-tolerant resource estimate for running Shor's algorithm against 256-bit elliptic-curve cryptography — the family that secures Bitcoin — claiming its cat-qubit-plus-qLDPC 'Walking Cat' architecture would need 19,397 physical qubits and 25.7 days. The figures are far below the millions-of-qubits estimates that dominated earlier surface-code analyses, and they rest on aggressive assumptions about biased-noise cat qubits and qLDPC overheads that IonQ's own hardware has not yet demonstrated. Treat it as a vendor-authored architecture study with explicit numbers to check rather than a measured capability; the numbers are the useful part, and they sharpen the post-quantum migration argument IonQ is also making.
AI & ML
Google threat researchers document a multi-agent framework harvesting thousands of credentials in six hours
Google Threat Intelligence Group says it observed a financially motivated crew running an autonomous, multi-agent attack framework that compromised thousands of credentials inside a six-hour window, alongside a broader pattern of attackers targeting proprietary AI systems. This is one of the more concrete data points so far on agentic offence moving from demonstration to operational use, with a named observer and a bounded timeline rather than a capability claim. The report does not disclose which models or agent stacks were used, so the technical bar the attackers actually cleared remains unclear.
The Hacker News — Autonomous AI Agents Compromise Thousands of Credentials in Under Six Hours
OpenAI claims a Millennium Problem result in fluid mathematics — verification pending
OpenAI says it has used AI to crack one of the seven Millennium Prize problems, the mathematics of fluid behaviour, according to Nature's news coverage. A claim at this level only counts once the mathematical community has worked through the argument, and Nature's own framing keeps the claim attached to the company rather than stating it as fact. Worth tracking precisely because the failure mode is well-established: previous OpenAI batches of claimed open-problem results have needed weeks of external scrutiny before their real scope became clear. No independent verification exists yet.
Nature — OpenAI claims huge maths breakthrough on a famed 'Millennium Problem'
Chipmakers converge on a process change said to lift High-NA EUV throughput 40%
Top chipmakers have coordinated on a change to how they use ASML's roughly $400-million High-NA EUV scanners, a shift reported to raise the machines' effective productivity by around 40 percent. Throughput, not resolution, has been the practical constraint on High-NA adoption, so an industry-wide alignment matters more than a single fab's tuning. The 40 percent figure originates with the toolmaker and its customers rather than an independent measurement.
Ars Technica — Top chipmakers embrace ASML's $400M machines, agree to crucial chipmaking change
Google's AI weather model gains accuracy by ingesting more raw satellite data
Google has updated its AI weather model to take in a wider set of raw satellite observations rather than relying as heavily on preprocessed analysis fields, and reports improved forecast accuracy as a result. The pattern mirrors how conventional numerical weather prediction improved — better assimilation of more observations — and suggests the learned models are converging on the same lever. Specific skill scores against operational baselines were not detailed in the coverage.
Ars Technica — Update to Google's AI weather model improves forecast accuracy
A startup bets that chips can recycle the energy they currently throw away as heat
MIT Technology Review profiles Hannah Earley, cofounder and CTO of Vaire Computing, which is building chips based on reversible computing — recovering the energy a conventional logic gate dissipates when it erases information rather than treating that loss as fixed. The approach targets the thermodynamic floor that increasingly bounds AI datacentre efficiency, and it is the same information-reversibility principle that underlies unitary quantum operations. This is a founder profile rather than a results announcement: no silicon benchmarks or energy-per-operation figures are reported.
MIT Technology Review — This founder is teaching chips how to recycle (their energy)