+ The US And UK Governments Both Stepped In On Anthropic's Most Powerful AI Model
On Monday, the US Treasury and the Federal Reserve pulled together an emergency meeting with Anthropic, the CEOs of five major banks, and federal cybersecurity officials. The subject was Claude Mythos, Anthropic's most capable AI model. Anthropic has kept Mythos out of the public's hands since February and only lets a small group of cybersecurity companies use it. The reason the regulators called the meeting: a test showed Mythos could find the weak points in a bank's network in under 12 minutes. Two of the banks at the meeting pulled their Mythos trials within 48 hours.
Four days later, the UK's AI Safety Institute published a 41 page report on the same model. The headline finding was that Mythos could plan and execute a complex cyberattack on a test environment with very little human help, and scored significantly higher than any previous AI on expert-level security tasks. Anthropic cooperated with both the US meeting and the UK report. CEO Dario Amodei said the reviews "reflect the kind of external scrutiny we have asked regulators to apply."
What it all means: This is the first time two governments have taken formal action on a single AI model in the same week. It is a preview of the regulatory cycle that will reach every major AI company by the end of the year.
+ OpenAI's Biggest Project Is Losing Key People
OpenAI lost three senior executives from its Stargate project to Meta last Friday. Stargate is OpenAI's $500 billion plan to build the data centers that will power the next generation of AI. The three who left were the people in charge of lining up the money, the partnerships, and the construction timeline. All three were hired away by Meta's new AI lab, run by the 28 year old founder of Scale AI.
That was not the only rough news. Two days earlier, the UK government paused its involvement in the UK side of Stargate. Norway's sovereign wealth fund pulled its soft commitment on Wednesday. And a UK minister told the Financial Times on Thursday that Stargate's power needs "are not compatible with our 2030 grid plan." The US side of the project, being built in Abilene, Texas, is still on schedule for Q3 according to Sam Altman.
What it all means: For three years, the story has been that the biggest AI companies can spend their way to the front of the pack. This is the first real crack in that idea. Building AI at this scale is harder than it looked, and competitors like Meta are already moving on the people who know how to do it.
+ Novo Nordisk Signed The Largest Enterprise AI Deal On Record
Novo Nordisk, the Danish pharmaceutical company behind Ozempic and one of the most valuable companies in Europe, announced on Tuesday that it will deploy OpenAI across its entire 71,000 person global workforce by the end of 2026. The five year contract is worth more than $1.4 billion according to two sources who spoke to Reuters, making it the largest single customer AI deal ever disclosed. The agreement includes a dedicated version of ChatGPT for Novo employees, custom AI models built on Novo's clinical trial data, and integration with the company's manufacturing and regulatory systems.
Novo's CIO said the decision was about internal productivity, not drug discovery. The company ran an 18 month test across three divisions before signing, and reported time savings of 11 to 19 percent on work like writing clinical trial documents and preparing regulatory submissions.
What it all means: Regulated industries are past the pilot phase. When one of the most cautious companies in the world signs a billion dollar AI contract, every other large pharma company has to explain why they are not doing the same.
+ Google Put An AI Agent Inside Chrome
Google launched AI Mode in Chrome on Thursday. The feature adds a Gemini powered assistant that can read any page, summarize it, compare it against other tabs you have open, fill out forms, and take multi step actions like booking a flight or reordering groceries. The rollout is immediate for all US Chrome users and will reach the UK, Canada, and Australia by the end of May. AI Mode is free, supported by ads on commercial pages, and does not require a Google subscription.
Chrome has roughly 3.4 billion users worldwide. Every one of them will see the AI Mode button within six weeks.
What it all means: Claude and ChatGPT both shipped their own AI browsers last year. Google has the distribution neither can match. By June the default way consumers research purchases, book travel, and compare prices will involve an AI that reads the web for them.
