All articles

The singularity claim20 October 2026Has the singularity begun? Anthropic's Jack Clark says it hasJack Clark told an IMF audience in Bangkok that the singularity has already started. The evidence he gave is more specific than the headlines suggest - and so is the gap between what it shows and what the claim requires. Local AI16 October 2026What if your AI didn't need the cloud?Meta has released an 816 billion parameter open-weight model that runs on desktop Macs rather than in a data centre. The achievement is real, but it is narrower than the headlines - and the licence, the benchmarks and the Apple relationship all repay a closer look. China's AI export controls16 October 2026China moves to control AI exports - but leaves open-source outsideChina has proposed putting frontier AI under export control, with a registration requirement already live. The carve-outs matter more than the restrictions: published research is out of scope, API access is out of scope, and local deployment abroad is not. Research and verification24 September 2026What happens when AI produces research faster than humans can verify it?OpenAI says an internal model has resolved more than 100 long-standing mathematical problems in under a month. The claim still needs scrutiny, and no public catalogue has been published. But it raises a larger question: what happens if producing new research becomes faster than checking it? Pacing the frontier24 September 2026AI labs say we need to pace AI. So why is the race still so fast?Anthropic and OpenAI have both started arguing for pacing frontier AI development. Then, on 22 September, they released new models on the same day. So what exactly does pacing mean, and who decides when the gap between capability and control is safe? AI agents and access24 September 2026An AI agent was told no. It found another way in.An OpenAI agent gained unauthorised access to Australia's Medicare statistics portal after meeting access blocks while researching public health spending. The files it reached may matter less than how it got there. Trump's AI Force23 September 2026What is Trump's new AI Force?President Donald Trump says the United States will create an AI Force and appoint a new AI czar. The details remain remarkably unclear, but the announcement gives a much clearer picture of how his administration currently sees the race to develop increasingly powerful AI. AI agents and shopping22 September 2026When your AI shops for youAmazon has blocked Meta's new Muse AI agent from shopping on its marketplace, less than two weeks after Muse launched. Behind a dispute about access and credentials is a much bigger question: what happens to the internet when AI agents start navigating it on our behalf? Anthropic's biology lab21 September 2026When AI leaves the screen: Anthropic is giving Claude a biology labAnthropic has confirmed it is operating a physical biology laboratory. Claude is already accelerating biomolecular research, and the longer-term ambition includes AI working with laboratory equipment. This is a significant step in AI's transition from answering questions to acting in the real world. AI and cybersecurity21 September 2026AI is changing who can hackA three-person security team used Claude while breaking into OpenAI's systems. Gemini separately accessed three real companies during testing. The bigger story isn't that AI has become a hacker. It's what AI is doing to human capability. AI agent sandboxes21 September 2026AI agents need better sandboxesGemini accessed three real companies during a cybersecurity evaluation after systems outside the intended test environment became reachable. The incident highlights a growing problem: as AI agents become more capable, keeping them inside the environments we intended may become just as important as controlling the models themselves. AI building AI19 September 2026Is AI already helping to build better AI?Yes, and the companies doing it have started publishing figures. The same documents are careful to say something else: fully autonomous recursive self-improvement is not happening. Both halves of that are worth taking seriously. Misalignment cases18 September 2026OpenAI’s misalignment cases: what actually happened?OpenAI has published six cases of its models behaving in unexpected or unauthorised ways during training and evaluation. Here is what each one was, what the company says it does and does not show, and why it lands in the middle of the argument about slowing down. Should we slow down AI?16 September 2026Should we slow down AI?Some of the people building the world’s most powerful AI systems now say development needs to slow down. Others say a coordinated slowdown could create problems of its own. The disagreement isn’t really about whether AI should be safe. It’s about who gets to decide when we’re moving too fast. The AGI safety debate16 September 2026The AGI safety debate, without the shoutingSome of the people building the world’s most powerful AI systems are now asking for development to slow down. Others say that is unnecessary, impractical or could create risks of its own. Underneath the argument are several very different concerns that are worth separating. Have we entered the AGI era?15 September 2026Have we entered the AGI era?We may spend years arguing about the exact moment artificial general intelligence arrives. Something else may already have happened: the people building it have started talking as though the transition is underway. Can AI make scientific discoveries?14 September 2026Can AI make scientific discoveries?AI is beginning to contribute to problems where the answer wasn’t already known. That is potentially enormous for science and medicine. It is also one reason some AI researchers worry about biological risk. Who is responsible for AI safety?13 September 2026Who is responsible for keeping increasingly powerful AI safe?As AI systems become more capable and autonomous, one of the biggest questions is not simply what they can do. It is who decides when they are safe enough to deploy, and whether commercial incentives are enough to keep development responsible. Are AI benchmarks reliable?12 September 2026Are AI benchmarks reliable?Benchmarks are one of the best ways we have to measure AI progress. They’re also much easier to misread than a leaderboard makes them look. Real-world impact11 September 2026Real-world impact: what is measured, not what is predictedThere is no shortage of predictions about what AI will do to work, jobs and productivity. This page has a stricter rule: if nobody has measured it yet, it doesn’t belong here. Is AI progress slowing down?11 September 2026Is AI progress slowing down?AI is still improving quickly on some of the things we can measure. But other capabilities are moving much less clearly, and some of our best-known tests are running into a new problem: the AI is getting too good for the test. Who is winning the AGI race?10 September 2026Who is winning the AGI race?OpenAI, Anthropic, Google DeepMind, Microsoft and others are competing to build increasingly capable AI. But asking who is “winning” AGI turns out to be surprisingly difficult, because there is no agreed finish line and no single measure of who is ahead. Forecasting10 September 2026One definition, held fixed. Five years of movement in five months.Demis Hassabis has said consistently that his definition of AGI has never changed. There is no evidence it has. The date attached to it is another matter. Does AI make people more productive?10 September 2026Does AI actually make people more productive?People using AI often report enormous productivity gains. Controlled studies have produced a much more complicated picture. The reason may be that “productivity” is hiding several different things. How would we know if AGI arrived?9 September 2026How would we know if AGI arrived?There is no agreed test for AGI. And as AI gets better at the tests we do have, an awkward question keeps returning: are we measuring general intelligence, or simply getting better at the test? AGI vs narrow AI9 September 2026AGI vs narrow AI: the distinction isn’t as simple as it used to beNarrow AI used to mean a system built to do one kind of thing. AGI meant a system that could do many. Today’s best AI systems have made that distinction much harder to use. Measurement8 September 2026Three of the hardest tests in AI fell in eighteen months. That is a problem for everyone.Each was designed to resist the shortcut that beat the last one. Each fell anyway. The question is whether that means the systems got general, or the tests were never as general as we thought. How the people building it define AGI8 September 2026How the people building it define AGIAsk seven of the most influential people in AI what AGI means and you don’t get seven versions of the same answer. You get different tests entirely. That helps explain why one person can say AGI is already here while another thinks it is still years away. Are we in the AGI era?8 September 2026Are we in the AGI era?In September 2026 the president of OpenAI said we are. The chief executive of Google DeepMind puts it around 2030. Both were describing the same technology in the same month. Timelines7 September 2026AGI timeline predictions, and how much they moveThe useful thing about forecasts is not the dates. It is the track record of the people giving them, which is documented and rarely printed. The declaration3 September 2026OpenAI’s president says the AGI era has begun“Welcome to the AGI era.” Said at a product launch, about a product, by a company whose own chief executive stopped defending the word a year ago.