All the AI that matters, in one place.
Search covers the labs, the trade press, the courts, and the regulators. Swap in a Programmable Search Engine before launch.
Anthropic, the company behind the Claude assistant, has set out the terms of its planned stock market listing. In an updated filing on Tuesday it said it would offer 825 million shares at $70 to $78 each, raising roughly $58 billion to $64 billion. At the top of that range the company would be valued at about $1.4 trillion, well above the $965 billion it reached in May but below the $2 trillion it was reportedly aiming for. Anthropic has also agreed to sell $26 billion of non-voting shares to chip and cloud companies, with Google and Nvidia each putting in $10 billion. The founders keep a special class of share carrying just over half the voting power. The price range is a useful gauge of how much confidence investors still have in AI after a shakier few weeks, and of how closely the tools many people use at work depend on a few very large companies.
TechCrunch — Nov. 10
New York is now enforcing one of the toughest state AI laws in the country: the biggest AI developers have to register, publish their safety plans and report serious incidents within 72 hours, or face fines of up to $3 million.
Claude can now join a video call as a visible participant and start drafting plans, timelines and action items while people are still talking, a sign of how quickly AI is moving into everyday office routines.
Hoax callers are using AI-generated voices and sound effects to make fake emergency calls convincing enough to empty schools and send out police, which puts real strain on emergency services.
A three-week hands-on test of the new AI assistants that check your email and accounts on their own found them mostly noisy. Read it before paying $20 to $100 a month for one.
A step-by-step guide to three new AI features in Safari on iPhone, iPad and Mac: sorting messy tabs, alerts when a web page changes, and building a browser extension just by describing what you want.
An honest test of free AI coding tools that run on your own computer: they managed a simple app but failed at debugging. It shows what a paid subscription still buys you and what kind of machine the free route needs.
One thing everyone is talking about, checked against what was actually said.
A researcher who spent three years on pretraining work at OpenAI and Anthropic resigned this week and posted that both companies are racing toward self-improving superintelligence. The headlines took the most alarming sentence and ran. Two things got left behind.
The first is that a departing researcher issuing a warning has become close to a genre. What made this one unusual is what happened next: Anthropic's own Alignment Science Lead, still at the company, agreed in public and put his estimate above one in ten within the decade. Senior staff almost never confirm a departing critic on the record.
The second is a line nearly every headline dropped. He said the risk from the models available right now is low, and that what concerns him is a future system improving itself faster than anyone can check it. Those are two different claims, and only one of them was made.
So, no, not the assistant on your phone — the people raising the alarm say so themselves. What they are describing is real, seriously contested inside the field, and some years out, and nobody involved claims it has been solved. Worth holding onto one caveat: a probability posted on social media is a considered guess, not a measurement.
Anthropic's price range is a reminder that big ambitions get tested once someone has to put a number on them. Most of us meet a smaller version of this at work: a proposal, a new tool or a project that arrives with a confident promise attached. This prompt helps you separate what the case actually shows from what it only hopes for.
Supplied as-is for general information. AI tools state wrong answers as confidently as right ones, so verify anything before you act on it. Nothing here is legal, financial, medical, or tax advice — for those, ask someone licensed to give it. Don't paste confidential, proprietary, or personal information into a tool whose terms let it train on what you type.