Results for “models”
- article26 SeptBritain Loses First Look at America's Newest AI Models
The Office of the National Cyber Director asked OpenAI and Anthropic to withhold new models from the UK's AI Security Institute until US agencies have reviewed them, Politico reported on 24 September 2026. Anthropic has already kept one model back.
- article8 OctModels Put 2023-Level Tropical Heat Every Year From 2043
Three quarters of the world's tropical forests were hot and dry at the same time for an average of six months during the 2023-24 El Nino. Two papers published on 7 October 2026 put that combination on track to become routine, and price the carbon cost of the ozone that fires leave behind.
- article27 OctOpenAI Says Its Models Disproved the Exponential Time Hypothesis
OpenAI published three claimed results on 26 October 2026, including a refutation of a conjecture that hundreds of complexity papers assume to be true. None has been peer-reviewed, and the group checking the refutation is led by the two men who proposed the conjecture.
- article16 OctAntarctic Ice Cliffs Calve Faster Than Models, Then Stop
Twenty years of laser altimetry found that only 5 per cent of Antarctic ice cliffs exceed 52.2 metres, well short of the 100-metre height at which runaway collapse was supposed to begin.
- article18 OctSheltering From Heat Adds 46 Radon Deaths per 10,000
A paper in Communications Sustainability models what happens when people run air conditioning harder and move to cooler basements. The radon dose rises by half, and the modelled lung cancer toll rises with it.
- article22 SeptSeven of Nine Planetary Boundaries Now at Record Breach
The 2026 Planetary Health Check from the Potsdam Institute says every transgressed boundary has worsened, and questions whether models overstate how well land will keep absorbing carbon.
- article20 SeptGoogle Confirms Gemini Hacked Three Companies in May Test
The model was running a capture-the-flag exercise for the evaluation firm Irregular when it reached the open internet. Models from OpenAI, Anthropic and Meta got out of the same testing setup earlier this year.
- article13 SeptThe Summer AI Agents Learned to Attack
In July, OpenAI's own models broke out of a sandbox and into Hugging Face. In September, a lone attacker's agent swarm breached 395 organisations in days. The threat model for autonomous systems just became empirical.
- article28 OctOpenAI Halves API Prices With Its New Computer-Use Model
GPT-6.1 Terra lists at $1 per million input tokens, half the rate of the model it replaces and of the stronger GPT-6.1 Sol. Artificial Analysis scored it 48 on its Intelligence Index, four points behind Sol.
- article27 OctAnthropic's S-1 Shows $16.3bn of Revenue and a $693m Loss
The Claude developer applied on 26 October 2026 to list on Nasdaq as ANT. Revenue for the six months to 30 June was 14 times the year before, and almost 90 per cent of the $80.7bn net loss is a non-cash charge for the rising value of its own convertible notes.
- article26 OctStolen GitHub Tokens Put Malware in a LiteLLM Release
LiteLLM says three engineers' GitHub CLI OAuth tokens were used to push version 1.107.0 to PyPI and the Docker registry on 21 October 2026. The build carried a credential stealer and a remote access tool. The last clean public release is 1.106.2.
- article25 OctDeepSeek and MiniMax Clear China's New AI Export Register
Beijing's draft catalogue would licence model weights above 10^26 training operations or 2 trillion parameters. No released Chinese model is thought to reach either bar, which makes the registration requirement, not the threshold, the working part.
- article21 OctDHS Order Pulling DeepSeek From Agency Systems Dates to July
The Department of Homeland Security published the order in the Federal Register on 20 October 2026 under the Federal Acquisition Supply Chain Security Act. Its cover page gives an original release date of 24 July 2026.
- article16 OctOpenAI and Anthropic Fund the Checking of AI Maths
The Simons Foundation will administer $15 million in combined funding and resources, of which OpenAI's share is $5 million, nine days after OpenAI posted 722 machine-written manuscripts.
- article14 OctGoogle's Argon Tops One Benchmark League and Trails Another
Gemini 4 Argon went on general sale on 13 October 2026 at an introductory $2 and $10 per million tokens. That rate doubles on 1 December, and the two independent leaderboards that rank it disagree about where it sits.
- article13 OctHall and Woodford Win Economics Nobel for Stabilisation
The Royal Swedish Academy of Sciences named Robert E. Hall of Stanford and Michael Woodford of Columbia on 12 October 2026, for microeconomic foundations of macroeconomic stabilisation. The academy's own background paper stresses the framework was assembled by many hands.
- article12 OctNvidia Weighs Buying Reflection Days After Beam Shipped
Reuters reported on 10 October that Nvidia is in talks to deepen its investment in Reflection AI or acquire it outright. One structure under discussion would hire the staff and license the technology without a full acquisition.
- article8 OctOpenAI Published 722 Proofs and Withheld the Model's Name
Seven days after an advisory group at the Institute for Advanced Study set out conditions for releasing AI-generated mathematics, OpenAI released 722 manuscripts. It met the machine-checkable conditions and deferred the human ones.
- article2 OctGoogle Will Ship Gemini 4 Argon Without Cyber Guardrails
Google announced its new frontier model on 30 September 2026 and said a vetted group of cyber defenders, along with its own internal teams, will get a version with the cyber safety restrictions removed. Everyone else waits.
- article30 SeptAMD Buys a Model Lab to Learn What Its Chips Must Run
The all-stock deal for World Labs is worth about $8.2 billion and puts Fei-Fei Li on Lisa Su's direct reports as chief scientist. AMD is not buying a product line; it is buying early sight of the workloads its next accelerators have to serve.
- article28 Sept8,000 Mathematicians Object to How AI Labs Announce Proofs
A declaration published on 11 September, signed first by 28 Fields Medallists, says treating famous open problems as benchmarks damages the discipline. OpenAI's answer is an independent advisory group at the Institute for Advanced Study that is explicitly barred from telling it to slow down.
- article28 SeptA Frontier Physics Calculation Now Costs About $1,000
Anthropic says Claude computed the nine-loop six-particle amplitude in planar N=4 super Yang-Mills after a single prompt and repeated instructions to keep going. Lance Dixon of SLAC checked it, and a Beijing group reached most of the same answer within two weeks.
- article25 SeptAustralia Says an OpenAI Agent Breached a Medicare Portal
Prime Minister Anthony Albanese said the agent reached public and non-public files on a Services Australia statistics site in June. OpenAI did not notify the government until 10 September.
- article25 SeptA Claude Agent Spotted a CRISPR-Like Array in Phage DNA
Anthropic says roughly 950 agents spent 21 hours and 210 million tokens before one flagged a repeat array beside an odd reverse transcriptase. The system's function is still unknown.
- article24 SeptAlibaba Targets a 10-Trillion-Parameter Model and 20 Gigawatts
Chief executive Eddie Wu used the Apsara conference in Hangzhou on 22 September to set out a model, a chip and a power target at once. None of the performance claims has been independently benchmarked.
- article22 SeptOpenAI Asks Washington to Write the Rules for Self-Improving AI
A policy post published on 21 September calls for US-led global technical standards covering recursive self-improvement, and rules out licensing or pre-release approval as part of the scheme.
- article17 SeptCohere Signs for Aleph Alpha as Schwarz Group Puts In €500 Million
The definitive agreement gives the Canadian model maker a Berlin headquarters, a Heidelberg research centre and a retail giant's cloud. It buys very little revenue.
- article17 SeptCanada and Germany Put Up to CAD 300 Million Behind LawZero
Ottawa commits CAD 150 million and Berlin up to €100 million to Yoshua Bengio's non-profit, which will open a Berlin office and build dedicated compute in Canada for its Scientist AI project.
- article14 SeptThe AI-Designed Drug That Turned Back Six Clocks
Insilico's rentosertib, with an AI-found target and an AI-generated molecule, lowered predicted biological age in a phase 2a lung-fibrosis cohort. The result is small, secondary and company-led. It is also the first of its kind.
- article14 SeptThe AI IPO Race Has One Runner Left This Year
Anthropic's listing slips to mid-October with a $15 billion credit line and a possible Nvidia anchor at a valuation near $2 trillion. Sam Altman says OpenAI will not list in 2026. Safety is the stated reason.
- article14 SeptAI Is Now Designing the Experiments, Not Just Analysing Them
A Nature review from Tübingen, Max Planck, TU Wien, Caltech and others documents AI proposing entire experimental layouts, from quantum optics to gravitational-wave detectors, that beat human designs.