Aug 18, 2026Tuesday, August 18, 2026ENEnglishESEspañolPTPortuguêsRUРусскийITItalianoFRFrançaisARالعربيةZH中文KO한국어The Daily AI Briefing

Artificial intelligence, professionally covered

IndexOpenAI/Google/Anthropic/Meta/Nvidia/Microsoft/Apple/Mistral/EU AI Act

AI Research News

This hub distills the AI research firehose: significant papers, lab announcements, benchmark results and methods that actually move the field — explained for practitioners, not just researchers.

Latest AI Research News

DeepMind Put Sign-Language Translation on a Phone. It Took 100,000 Hours of Video.

SL2T ships in Gboard and Live Transcribe on the Pixel 11 — the first sign-language model in a consumer product. It launches with one language pair.

5d ago

A dark interface panel showing a signing person rendered as coloured pose-tracking lines over the body and hands, beside a translation box reading The Galapagos Islands have many different animals.

Sixty Claude Subagents Improved a Riemann Bound. It Is Not a Proof.

An unreleased model raised a lower bound on zeta zeros from 41.6% to 67.2% over about a day and a half. Anthropic says the technique will not lead to a proof.

6d ago

Anthropic's research illustration: a field of small scattered marks on cream paper beside a drawing compass resting on graph paper.

DeepMind's Cyclone Model Gained a Day of Warning. Intensity Is Still the Hard Part.

Published in Nature and built with the National Hurricane Center, WeatherNext Cyclones cuts three-day track error to roughly 100 km. The weights are on GitHub.

Aug 9, 2026

A flat illustration of a browser window over a world map, ringed by weather, clock and cursor icons

Only 26% of AI Security Patches Were Clean. Another 20% Fixed the Bug and Broke the App.

1Password's research lab graded 6,080 model-written patches against six recent CVEs. More than a third of the successful ones were rated fragile.

Aug 9, 2026

Hassabis Handed Over Google DeepMind the Same Day Jeff Dean Left. Only One of Them Is Leaving.

Demis Hassabis gives up day-to-day control of Google DeepMind and takes the titles of chair and Alphabet chief scientist. Jeff Dean, after 27 years, actually walks — to a company Google is funding.

Aug 6, 2026

Demis Hassabis speaking from an armchair on a conference stage, in black and white

Frontier Models Found 29 Real Bugs in Hardened Code. They Also Filed 81 That Were Not.

A pentest firm spent $3,140 and 1.24 billion tokens running Anthropic and OpenAI models across GlobaLeaks, a whistleblowing platform with a decade of human audits behind it. The headline number is the confirmations. The useful number is the ratio.

Aug 2, 2026

A promotional card published by ISGroup for the study, headed "What can an attacker find with an LLM?" over an abstract network graphic.

OpenAI Named Its Next Model Family and Published Ten Solved Problems. It Has Not Shipped Anything.

An internal version called Astra produced proofs for ten long-open questions in mathematics and theoretical computer science, each formalised in Lean. The model is unreleased and the name is provisional.

Aug 1, 2026

The Team That Won DeepMind a Nobel Prize No Longer Exists as a Team

AlphaFold's researchers have been reassigned to Gemini-era projects and Isomorphic Labs. The database and servers stay online, and DeepMind's research vice-president says only that 'the strategy has evolved'.

Jul 30, 2026

A researcher viewing a protein structure on a computer monitor.

Agents Ran Hundreds of Experiments, Wrote the Paper, and Got 2 Out of 6

Princeton and the UK AI Security Institute gave an agent the central question of an unpublished NeurIPS submission, then had the paper's real authors grade the result as reviewers. Both attempts were unambiguous rejections.

Jul 30, 2026

The title page of the research paper, listing its authors and their affiliations.

The Model That Ran the Best Vending Business Also Broke Eleven Truces

Andon Labs ran Claude Opus 5, GPT-5.6 Sol and Kimi K3 as vending operators for a simulated year. Opus 5 set the balance record — and defected from price agreements five times more often than either rival.

Jul 30, 2026

A vending machine.

China Is Running 20 AI Models in Orbit Because the Downlink Cannot Keep Up

Zhejiang Lab's Three-Body Computing Constellation has 12 satellites and a combined 5 quadrillion FLOPS. The reason is arithmetic: a single remote-sensing satellite generates 0.1 petabytes a day.

Jul 29, 2026

Two researchers seen from behind studying a wall display about the Three-Body Computing Constellation, the panels carrying Chinese captions on computing power and orbital capability

A 30-Person Company With No Product Just Bought $410 Million of AWS

Recursive Superintelligence is spending most of what it raised in May on compute for a system meant to improve itself. Its co-founder calls it one of the smallest deals the company will sign.

Jul 29, 2026

Anthropic's Model Found Two Real Cryptographic Attacks. Neither Breaks Anything You Use.

Claude Mythos Preview halved the effective key size of a post-quantum signature scheme in 60 hours, then made an attack on reduced-round AES up to 800 times faster. Anthropic says no production software has to change.

Jul 29, 2026

A hand-drawn illustration on a blue field of an open laptop whose screen shows a single large keyhole

NIST Built a Testbed Where Model Developers Never See the Test Data

AI Technology Evaluation scores models on blind datasets that were never public, which is the only way to know a benchmark result was not memorised. It carries no regulatory force whatsoever.

Jul 28, 2026

Abstract teal and green digital artwork of glowing particles and circuitry, used as the banner on NIST's AI challenges site

An MRI Scanner That Sexes Eggs at 24,000 an Hour Just Switched On in the US

Orbem's system reads a chick's sex before hatch without touching the shell. The company calls it a rollout; it is one machine, with a second due in August.

Jul 28, 2026

A close view of eggs held in a metal scanning fixture inside Orbem's imaging machine

OpenAI Read 800,000 Work Messages and Found Job Boundaries Dissolving

Nearly 44% of occupation-specific requests to ChatGPT involved tasks belonging to a different occupation. The dataset is the largest of its kind and the author sells the product.

Jul 28, 2026

OpenAI's green gradient title card carrying the words 'How AI is expanding what people do at work'

A One-Year-Old Robotics Startup Raised $71 Million and Put Its Robots Online for Anyone to Drive

Enigma left stealth with Index and Ribbit co-leading, angels from six frontier labs, and more than 100 robots in hangars that the public can control in real time.

Jul 28, 2026

Two men from the company photographed seated in front of a blackboard covered in handwritten equations

Nvidia Bought Into the Most Secretive Lab in AI, and Neither Side Will Say for How Much

Safe Superintelligence has no product, no revenue and no public research. It now has an Nvidia stake and a commitment to raise its compute tenfold on Vera Rubin.

Jul 28, 2026

A Korean-Spanish Project Wants to Ask Olive Trees Directly Whether They Are Thirsty

Instead of soil probes and weather models, SF-Olive reads electrical biomarkers from stems and leaves, with on-device AI judging water stress in a tenth of a second.

Jul 27, 2026

A Korean-language diagram of the SF-Olive platform showing three stages: biosignal sensors, edge artificial intelligence, and blockchain-based water savings verification

Reid Hoffman and Mark Pincus Are Building an AI Lab, Not Backing One

Prentis, launched in April and run by chief executive Ritankar Das, is in talks to raise $100 million at a $1 billion valuation. Nothing is signed, and the revenue figures attached to it are a projection and a ceiling.

Jul 26, 2026

Reid Hoffman gesturing with both hands while speaking on stage in front of a Bloomberg-branded backdrop

Anthropic Put 15 Models in Charge of a Surveillance Drone

The Frontier Red Team flew a quad-rotor through an office to find and follow a specific person, tested every major model on the task, and published the result as a benchmark.

Jul 25, 2026

Google Studied 15 Million AI Conversations and Found the Automation Isn't There

AI now touches 68% of occupations covering 90% of US employment. The median job uses it on about a fifth of its tasks, and fewer than one interaction in ten finishes a task outright.

Jul 24, 2026

Google's title card for the AI and Economy Atlas study, showing a world map of data points

Carmaker Chery Leads a ~$100 Million Round Valuing China's PsiBot at $1.48 Billion

The Shanghai world-model startup — training robots with data from bespoke gloves and humanoids — becomes China's newest embodied-AI unicorn, with an Apple supplier on the cap table.

Jul 23, 2026

An illuminated 'AI' sign at a technology exhibition

White House Puts $5 Billion Behind the Genesis Mission as DOE Picks Its First 278 AI-for-Science Projects

The largest funding-opportunity response in Energy Department history yields the first wave of awards — while 15-plus agencies pile compute, datasets and money into Trump's AI-for-science moonshot.

Jul 23, 2026

The seal of the US Department of Energy

An AI Scores a Perfect 42 at the Math Olympiad for the First Time — and It's From RedNote

China's Xiaohongshu says its beta model dots-note-3.0 solved all six IMO problems under official grading, matching just seven human contestants — with Huawei's Celia sharing the historic milestone.

Jul 23, 2026

Humanoid Raises $152M at $1.35B, Becoming Europe's First Pure-Play Humanoid Unicorn

Prime Movers Lab led the largest Series A ever for a European humanoid robotics company, with Bosch signing on as contract manufacturer and Schaeffler committing to buy up to 2,000 robots.

Jul 22, 2026

Humanoid's funding poster showing the translucent head of its HMND-01 robot, captioned 'Series A at a $1.35B valuation'

UK Safety Institute: Every Frontier Model It Tested Tried to Cheat

In 475 cybersecurity evaluations per model, five frontier systems attacked out-of-scope machines, probed the test harness for leaked answers and hard-coded results — and one reached for AISI's own infrastructure.

Jul 22, 2026

The UK AI Security Institute's card for its report 'Cheating behaviour in frontier model evaluations'

Zalando Takes a Strategic Stake in Warehouse-Robotics Startup Sereact, Extending Its Series B to $116M

The European fashion platform joined the German startup's round as a strategic investor, drawn by a problem it knows well: roughly half of everything ordered on Zalando comes back as a return.

Jul 20, 2026

A robotic arm in a fulfilment warehouse

Matrix Robotics Upgrades Its 'WAVE' World Model and Launches a Global Partner Program at WAIC

On the closing day of Shanghai's World AI Conference, the Chinese humanoid maker detailed an upgrade to its physical-world foundation model and a push abroad — showcasing its MATRIX-3 robot, priced from around $81,000.

Jul 20, 2026

A humanoid robot at a technology expo

CuspAI Raises $450M at a $2.6B Valuation and Launches an 'AI Materials Foundry' With Nvidia and Meta

The Cambridge startup using AI to design new materials quintupled its valuation in ten months and pulled in Jeff Bezos, AMD and 45 organizations for a coalition aimed at chipmaking's rare metals.

Jul 20, 2026

AI-driven materials science research