AI News
Short, sourced, and free of hype. Page 2 of 16.
The Same Model Scored 30 Percent Alone and 100 Percent Inside Nvidia's Agent System
Nvidia's AVO cleared all 183 levels of ARC-AGI-3 using Claude Opus 5, a model that scores about 30 percent on its own. The interesting part is not the score, it is what changed.
22 Aug 2026
Nvidia Paid 6 Billion Dollars for Poolside's Software and 109 of Its People, and Calls It a Licence
No acquisition, no acquihire. Nvidia licenses the platform, invests another billion, and hires most of the team, while Poolside stays independent on paper.
22 Aug 2026
A Frontier-Class Model Appeared With No Name on It, and It Is Free Until Next Week
Ox Alpha turned up on OpenRouter with a million-token context and no owner. Community fingerprinting points at Zhipu's unreleased GLM, but nobody has claimed it.
22 Aug 2026
Adobe Firefly Can Now Make the Music, the Voiceover and the Sound Effects
Adobe moved Generate Music, Generate Speech and Generate Sound Effects out of beta on 20 August. All three come cleared for commercial use, which is the part that actually matters.
21 Aug 2026
This Robot Learns a New Task From One Short Video, No Training Required
Generalist AI's GEN-1.5 takes a three to twelve second demonstration as a prompt and performs the task at 59 percent success, rising to 83 percent after five minutes of data. All numbers come from the company.
21 Aug 2026
Google Adds a Button That Lets Readers Pick Their Own Favourite Sources
Publishers can now embed a Preferred Sources button on their own site. Google says readers are twice as likely to click through to a source they have marked as preferred.
21 Aug 2026
Nobody Passed: The First Report Card on How AI Labs Police Their Own Systems
Guidelight graded Anthropic, OpenAI, Google, xAI and Meta on six basic internal control practices. The best score was a C plus, Meta got an F, and the assessment used only public documents.
21 Aug 2026
OpenAI Wants to Catch Misuse Without Ever Seeing Your Data
OpenAI previewed Private Safety Processing, a system meant to spot abuse patterns across several conversations while keeping zero data retention for business customers. Anthropic still requires 30 days of logs for its strongest models.
21 Aug 2026
A Third of the New Web Shows Signs of Being Written by AI, Pew Finds
Pew Research scanned roughly half a million web pages and found AI authorship signals in 35 percent of those published after ChatGPT launched. Commercial sites score ten times higher than universities and government.
21 Aug 2026
Reinforcement Learning Pioneer Rich Sutton Calls Synthetic Data a Big Mistake
Sutton argues that a world of endless complexity cannot be learned from data a model generates for itself. It is a direct challenge to one of the industry's favourite shortcuts.
21 Aug 2026
AI Could Write Like a Human. Safety Training Is What Gives It Away
Pangram's CTO argues that language models are detectable not because they lack ability but because post-training narrows how they express themselves. The effect has a name: mode collapse.
21 Aug 2026
Amazon Just Made Its AI Assistant Free on Fire TV, and Quietly Dropped a 19.99 Dollar Fee
Alexa+ is rolling out to all compatible US Fire TV devices at no cost, with no Prime membership required. The upgrade is automatic. The catch is that the free tier stops at the television.
20 Aug 2026
Anthropic Left Claude Alone With 15 Protein Targets for 48 Hours, and the Lab Results Came Back
Claude designed protein binders against 14 of 15 targets in wet lab tests run by Adaptyv Bio and Twist Bioscience. Hit rates ran 22 to 35 percent against a typical 10 to 15. On one target it beat a competition winner.
20 Aug 2026
One Link Was Enough to Empty Your Copilot, and the Fix Took Eight Months
Microsoft patched CVE-2026-24301 on 18 August. Varonis calls it CoSnitch: a single click on a crafted link made Copilot read a victim's Gmail, Drive and Calendar and send the contents to a stranger.
20 Aug 2026
Europe's Next AI Data Centres Are Being Built 175 Kilometres From Anywhere, Because That Is Where the Power Is
JLL figures show Europe's coming AI data centres will sit an average 175km from major hubs, up from 46km for 2022 to 2025 projects. Powered land in Amsterdam costs 2.36 million euros per megawatt against 512,000 elsewhere.
20 Aug 2026
The FDA Has Cleared 1,357 AI Medical Devices. Three Were Tested on Whether Patients Got Better
A PLOS Digital Health analysis found that of 1,357 FDA-authorised AI medical devices, 34 appeared in registered clinical trials and only 3 measured outcomes like mortality or hospitalisation.
20 Aug 2026
An Oxford Spinout Sold Anthropic 250 Million Dollars of Chips That Do Not Exist Yet
Fractile is reportedly raising 600 million dollars at a 6.5 billion valuation, six times its May mark, after an initial deal to supply Anthropic with inference chips. The chips are not expected until 2027.
20 Aug 2026
Google Is Giving Students a Free Year of Its Paid AI, and the Real Product Is the Study Tools
Eligible US college students get 12 months of Google AI Pro, normally 19.99 dollars a month. Students in 140 other markets get AI Plus. Alongside it, Google launched a Student Hub with study notebooks and practice quizzes.
20 Aug 2026
Watching Its Own Model Now Costs OpenAI 20 Percent Extra, and It Paused a Training Run to Do It
OpenAI says safety monitoring adds roughly 20 percent to the inference compute being monitored. It paused reinforcement learning for two weeks and its largest planned frontier run is still on hold.
20 Aug 2026
Give a Frontier Model Only a Bibliography and Ask for the Idea. It Gets There 3 to 15 Percent of the Time
A new benchmark called Reconstruction strips away everything except a paper's reference list and asks models to recover its core finding. Seven frontier models scored between 3 and 15 percent. A multi-agent tournament reached 42.
20 Aug 2026
Your Next PC Upgrade Just Got Five Times More Expensive, and AI Is the Reason
DDR5 memory prices are up nearly 500 percent in twelve months. A 128GB kit that once bottomed out at 329 dollars now lists at 3,399. The cause is not scalping, it is AI data centres eating the wafers.
19 Aug 2026
Anthropic Is Reportedly Paying 7 Billion Dollars for Software That Makes Chips Go Further
Anthropic and Decart are exchanging advanced drafts of a 7 billion dollar acquisition, mostly in shares, with signing targeted for early September. Decart claims models can run up to 8 times faster on the same hardware.
19 Aug 2026
OpenAI Cut Its Flagship Model to Half Price, and the Reason Is Sitting on a Leaderboard
GPT-5.6 Sol is 50 percent off through 18 September on OpenRouter and Vercel's gateway, taking it to 2.50 dollars per million input tokens. It follows cuts to Luna and Terra as Chinese models gain ground.
19 Aug 2026
A Legal AI Company Built Its Own Model, and It Started From an Open Chinese One
Harvey launched Harvey II with a Memory feature that learns each lawyer's drafting style, plus Tenet, its first in-house legal model, post-trained from a version of Moonshot's open-weight Kimi K3.
19 Aug 2026