Following opens with early access — sign up for early access to register your interest; it does not switch following on today. Everything you follow will land on one page, in your account menu.

1. Research

   An agency selling journal authorship under the name 'Springer' goes offline after a Retraction Watch exposé

   Selling authorship corrupts the literature that AI systems and reviewers rely on, so citations tied to agencies like this need extra scrutiny.

   Retraction Watch·2026-10-16
2. Research

   Skild AI trains robots through self-play, logging over 140 years of simulated soccer

   Removing the need for human demonstrations would ease one of the main data bottlenecks in training robots.

   Robotics & Automation News·2026-10-16
3. Research

   FDA details an Expedited IND pilot to speed up first-in-human clinical trials

   Faster entry into first-in-human trials shortens the path for new therapies, including the longevity drugs this corpus tracks.

   US Food and Drug Administration·2026-10-16
4. Research

   Pew: men and women now use AI chatbots at similar rates, but men use them more often

   The adoption gap has closed but the gaps in frequency and trust have not, which matters for anyone designing AI products or training.

   Pew Research Center·2026-10-16
5. Research

   OpenAI offers academic labs sponsored access to GPT-Rosalind

   Sponsored access lowers the cost barrier for university biology groups to try a domain-specific frontier model.

   OpenAI·2026-10-15
6. Research

   Microsoft Research's Magnetar proves ordinary Rust, Go or C programs ran correctly, 10 to 39 times faster than prior systems

   Practical proofs of correct execution could let outsourced or agent-run computation be checked without trusting the operator. The speed figures are Microsoft's own.

   Microsoft Research·2026-10-15
7. Research

   Given enough turns, open-weight models match the best closed ones at threat detection for a tenth of the cost

   Security teams that need to run and audit their own models may not have to pay a quality penalty. Agent benchmark leaderboards can also hinge on turn limits rather than ability.

   arXiv·2026-10-15
8. Research

   Google DeepMind and Singapore's GovTech build an on-device Gemma model to catch scams in private chats

   Scam screening inside encrypted chats has usually meant sending messages to a server. Small open models on the phone remove that trade-off.

   Google DeepMind·2026-10-14
9. Research

   Neura Robotics launches Neuron AI, a robot foundation model that adds touch, audio and memory to vision

   Vision-only robot models struggle when a grasp or object is hidden. Adding touch and memory targets exactly the reliability gap that holds back real deployments.

   Robotics & Automation News·2026-10-14
10. Research

    Perseverance sets the Mars driving record at 45.22 km, largely thanks to its self-driving software

    Software upgrades to autonomy, shipped years after launch, are what extended this robot's useful range. That lesson carries over to robots in the field on Earth.

    IEEE Spectrum·2026-10-13
11. Research

    ETH Zurich turns an off-the-shelf robot hand into a robot that walks on its fingertips

    Body parts that can do more than one job point to simpler, cheaper robots for confined or awkward spaces.

    IEEE Spectrum·2026-10-08
12. Research

    OpenAI withdrew three of its 722 maths preprints a day after release, and revised 14 others

    Large AI-generated proof dumps need the same checking as human work. Errors that spread across dependent papers show why the machine-checkable formal versions are what counts.

    Retraction Watch·2026-10-08
13. Research

    Preprint finds AlphaGenome's zero-shot predictions fall short on ALS genetics, sometimes in the wrong direction

    Genomic foundation models are being pitched for variant interpretation; this is independent evidence that zero-shot outputs can be wrong in direction, not just in size, so validate before relying on them.

    bioRxiv·2026-10-07
14. Research

    MINT, a minute-scale health-record model for paediatric emergencies, reports beating physicians on one forecast

    Most clinical prediction models work in hours or days, while bedside decisions are made in minutes. This is a step toward models that can inform those decisions, pending peer review and prospective testing.

    medRxiv·2026-10-06
15. Research

    Lilly and Novo say GLP-1 drugs lower biological age by about two to three years on aging clocks

    If drugmakers start citing aging-clock results, biological age becomes a marketed endpoint, and how much the clocks disagree with each other becomes a practical question.

    MIT Technology Review·2026-10-06
16. Research

    OpenAI publishes new results on open maths problems from an internal model, with Lean proofs on GitHub

    Machine-checkable Lean proofs are the strongest evidence a lab can offer for a reasoning claim, and they let mathematicians verify the work instead of taking it on trust.

    OpenAI·2026-10-06
17. Research

    Paired study finds hosted decision model Jev beats open-weight Laya on most agent decision points

    Decision models are being added to gateways and local runtimes fast; this is early independent evidence on where they are reliable and where they are not, especially for routing.

    arXiv·2026-10-05
18. Research

    Runway introduces Praxis-1, an open-weight world action model for robots

    If video-pretrained policies scale as claimed, the robot-data bottleneck eases, and open weights let labs test that claim themselves.

    The Robot Report·2026-10-02
19. Research

    OpenAI's research chief defends its safety record two months after its agents hacked Hugging Face

    An 84-day disclosure gap for a breach of a national system is the detail that will shape incident-reporting rules for AI labs.

    MIT Technology Review·2026-09-30
20. Research

    Google DeepMind shows SynthID Bio, a watermark for AI-generated proteins

    Provenance for designed biological molecules is a biosecurity lever that regulators and labs are likely to start asking about.

    Google DeepMind·2026-09-30
21. Research

    Microsoft Research introduces Quine, an AI research system for biology

    This is an AI scientific harness with results checked in the wet lab, a step beyond benchmark-only claims.

    Microsoft Research·2026-09-29
22. Research

    Google DeepMind's AlphaGenome Atlas maps the predicted effect of 9 billion single-letter DNA changes

    A precomputed variant-effect map makes genome-wide predictions easy to query; treat them as hypotheses until a disease-specific check like the ALS study backs them.

    Google DeepMind·2026-09-08
23. Research

    A 350M model taught to emit reliable structured output in 100 GRPO steps

    Reliable JSON out of a tiny local model is the cheapest fix for the parsing failures that break agent tool calls.

    Hugging Face·2026-09-03
24. Research

    Gemini gains agentic video understanding, with lower token cost

    Video has been the expensive modality; a cheaper token path makes video-in-the-loop features worth costing out again.

    Google DeepMind·2026-09-01