Analysis
What the big stories mean, with our own charts and every source linked.
TechGoogle keeps its best model for the top plan, Claude learns spreadsheets, and a new music model edges past SunoGemini 4 Carbon is out, but in the Gemini app only Ultra subscribers get it. Claude picked up spreadsheets and whiteboards. Apple appears to be building a way to join Macs into one AI machine. And a music model from a small studio tied Suno at the top of a blind test.
Tech948 ways to root: OpenAI's agents break Linux at scale, Perplexity hires out teammates, and a voice model fits in 32 MBAn OpenAI pipeline reproduced 1,724 Linux kernel bugs and wrote working root exploits for 948 of them, then drafted the patches. Agents also moved into team chats and into Cursor's deploy button. And at the small end, a speech model shrank to less than a few phone photos.November 9, 2026
TechClaude Fable 5.2: cheaper, faster, and twice as good at real remote work, says AnthropicAnthropic's new top model costs a fifth less per output token than the last one and doubles its score on a test of paid freelance jobs. It also ran a team of 1,000 agents that formalised Fermat's Last Theorem in about 30 hours. The figures are Anthropic's own, and for most work it still recommends a cheaper model.November 9, 2026More analysis
- TechWe re-ran the test that said Qwen writes worse code for Americans. We could not confirm itA security firm reported that Alibaba's coding model wrote more hackable apps when told the customer was a US defence contractor. We repeated the experiment with the same benchmark and model on 1,090 apps. One finding failed to repeat, one stayed open, and one was too small a sample to judge.October 28, 2026
TechWho is the code for? Qwen writes weaker apps for a US customer, AI attackers move in three hours, and Claude proves its fixesA security firm found that the same coding model wrote more exploitable apps when told the customer was a US defence contractor. Microsoft caught a state-linked group running an AI pentest kit against thousands of companies. And on the other side, models started proving that code is safe.October 27, 2026
TechWe ran Stanford's new 'whole draft first' AI model ourselves. It aces Sudoku and loses numbersNext-Concept Modeling writes a rough version of the whole text first, then sharpens it level by level. We rented a $0.34-an-hour graphics card and reran the published models. The paper's results held up. So did its warning: names and numbers get lost.October 23, 2026
TechWhen AI writes the code, where does human value move?Producing a first draft is getting cheaper. Understanding what is needed, choosing what to build and taking responsibility may become more valuable. A view from inside IT on what we will be paid for, how the next generation will learn, and what AI should and shouldn't do in schools and literature.October 15, 2026
TechTreat the model like an insider: Anthropic's agents misbehave, Nadella writes the rules, and AI learns to answer yes or noAnthropic reported that its agents exploited website flaws during tests. Satya Nadella said frontier models should be treated as insider risks, with an emergency brake. And the week's money went to models that don't write at all: they only decide.October 12, 2026
TechThe week the guards went up: Apple locks the disk, OpenAI loses its safety voicesApple is making it harder for apps, and the AI agents inside them, to read everything on a Mac. OpenAI lost three safety researchers and a safety lead in one week. And the tools for keeping agents in line got smaller, cheaper and more personal.October 5, 2026
TechFree Gemini shrinks to one small model on 9 October. What to do if you use it for workFrom Thursday, people who use the Gemini app without paying get only Flash-Lite, Google's smallest model. Flash and Pro move behind a subscription. The free lunch in AI chat is getting thinner, and it pays to know what you actually need.October 4, 2026
TechOpenAI wants ChatGPT to be your office. What DevDay means for workAgents that keep working when you close the tab, a shared space for teams, apps that run inside the chat and a shop for business software. OpenAI's biggest DevDay was less about models than about where work happens.October 4, 2026
TechHalf the people on the call thought it was human. It wasn’tA start-up called Tavus says its new video model fooled 48% of people in a one-minute call. The same week, Ideogram released an image editor that can change a photo again and again without leaving a trace. What we see on a screen is getting harder to trust.October 2, 2026
TechGoogle's Gemini 4 Argon: a frontier model at a mid-range price, but not for you yetGoogle says its new top model leads on coding, legal and finance work and cybersecurity, and it will cost no more per token than mid-range rivals at launch. First it goes only to vetted cyber defenders, as AI labs get more careful about who gets their most capable models.October 1, 2026
TechOpenAI's $500 plan and GPT-6.1 Sol: speed is the new premiumOpenAI now sells a $500-a-month ChatGPT plan built around raw speed, while its new mid-range model gets close to the flagship for a fifth of the price. Here is what changed, what it costs, and who should care.September 30, 2026
HealthAI is making cybercrime cheaper. Hospitals and small banks are the easiest targetsAttacks that once needed a team of skilled hackers can now be run by one person with an AI model. Here is what the numbers show, why local hospitals and community banks are exposed, and the simple defences that still work.September 29, 2026
TechWhy a chipmaker is paying $8.2 billion for an AI lab that builds virtual worldsAMD is buying World Labs, the start-up founded by AI pioneer Fei-Fei Li. The deal says a lot about where AI is heading next: from chatbots that write to models that understand space, and the robots and tools that need them.September 29, 2026
TechAnthropic's IPO filing, explained: huge growth, bigger bills, and a warning about humanityA draft of the prospectus for what could be the largest share sale in history shows how fast the AI business is growing, how much it costs to build, and what its makers say could go wrong.September 29, 2026
PolicyOpenAI built a model it decided not to release. Here's why that mattersGPT-6.1 Astra was meant to launch in October. It failed OpenAI's own safety tests on honesty and staying within its instructions, capping a summer of incidents that is changing how AI agents are built and regulated.September 29, 2026
TechThe week AI agents left the lab: voice, glasses, rogue swarms and a hardware watchdogAgents moved into phones, glasses and shipyards this week, while fresh disclosures showed how far test agents had already wandered. Nvidia's answer is to police them from a separate chip.September 28, 2026
PolicyCould AI go badly wrong? The AI safety debate, explainedA summer of rogue AI agents has pushed a niche argument into the mainstream. Here is what each side actually believes, what researchers and the public think, and a two-and-a-half-hour debate worth your time.September 28, 2026
PolicyMemory is China's AI bottleneck, and smuggling can't fix itNew estimates from Epoch AI show Huawei making a small fraction of Nvidia's compute, held back mostly by high-bandwidth memory. Smuggling helps at the margins, and trade data shows how quickly it dries up when the rules tighten.September 28, 2026
TechDeepSeek's $1 billion run rate, and what the number hidesChina's cheapest frontier lab is growing fast and raising at a $75 billion valuation. But a run rate is not revenue, prices are going up, and compute and a data probe now set the limits.September 25, 2026