Anthropic may raise its IPO share price after heavy investor demand, FT says

The Financial Times reports, citing sources, that bankers for Anthropic, the company behind the Claude AI, are weighing a higher price for its stock market debut. The reason given is strong investor demand. A higher price could lift the valuation target to $1.5 trillion or more. Separately, CNBC reports that Anthropic has opened a retail offer in Australia, so individual investors there can buy Class A shares ahead of its planned Nasdaq listing.

Google launches an AI service that picks from fixed answers and scores its confidence

Google has launched the Gemini Decisions API, powered by a new model called Gemini Decision 1. Instead of writing free text, it returns one of a set of predefined answers, such as yes/no or a choice, along with a confidence score. Google says it can evaluate up to 1,000 such decisions and calls them fast, calibrated and multimodal. The price is $0.10 per million input tokens.

Nvidia's board may remove CEO Jensen Huang after a second alleged affair, WSJ says

The Wall Street Journal reports, citing sources, that Nvidia's board is preparing to oust Jensen Huang, its founder and chief executive of 33 years, and met on Wednesday to decide his future. Nvidia shares fell more than 7% after hours. The move follows a New York Times report that Huang had an affair with Kimberly Powell, Nvidia's vice president of healthcare, who reports to him; the Times says five employees told it about the relationship. This comes days after Huang admitted an affair with a woman alleged to be a Chinese spy. House Republicans say Congress may subpoena him if he does not agree to a closed-door briefing, and Ars Technica reports the FBI is investigating.

First tests of AMD's Zen 6 server chips show them up to 93% faster than Zen 5

Tom's Hardware published its first in-depth look at AMD's EPYC 'Venice' server processors. They use AMD's new Zen 6 design and are made on TSMC's N2P process. In its tests, a server with two EPYC 9996 chips was up to 93% faster than one with two Zen 5 EPYC 9965 chips. The EPYC 9686F often beat Nvidia's Vera processor. AMD executive Madhu Rangarajan said the chips were not built specifically for AI agents, a dig at Nvidia and Arm, but still called them the best CPU for that work. Chief Zen architect Mike Clark said AMD did not see agent computing coming.

AI coding tool Grok Build is leaking data from over 35,000 apps, researchers say

Researchers found that Grok Build, an AI tool from SpaceXSI that writes apps from plain-language requests, is leaking data from more than 35,000 apps through shareable links, Ars Technica reports. The exposed data includes proprietary business information and other sensitive secrets, some of it belonging to SpaceXSI itself.

Amazon releases a free tool that runs AWS cloud services on your laptop, written by AI

Amazon Web Services released AWS Emulator, a free tool that simulates more than 200 of its cloud services on a developer's own machine, GeekWire reports. Existing tests, command-line scripts and coding agents can talk to it using the same AWS commands, without deploying to the cloud. AWS says thousands of autonomous AI agents wrote every line of its code and that its engineers did not write or review any of it. The tool is written in Rust, built from the same service descriptions the AWS SDKs use, and its source is on GitHub.

Nvidia scandal makes US chip export limits more likely this year, Politico says

Politico reports, citing sources, that Jensen Huang's affair has significantly raised the chances that Congress passes chip export limits this year through the National Defense Authorization Act, the yearly US defence law. Reuters found records showing that China's Northwestern Polytechnical University, which has close ties to the Chinese army, bought Nvidia H200 AI chips last month despite being on a US export blacklist. Axios reports, citing sources, that Commerce Secretary Howard Lutnick has stalled curbs on companies suspected of moving Nvidia chips to China, and that Treasury Secretary Scott Bessent confronted him over not blacklisting Nvidia customer Megaspeed; the Commerce Department denies that confrontation happened.

EU proposes a law to curb addictive app design and manipulative 'dark patterns'

The European Commission has published a draft Digital Fairness Act. It would ban manipulative 'dark patterns' such as fake 'almost sold out' claims, let people switch off endless scrolling, autoplay and constant prompts, and apply the same limits to AI companions and chatbots. It would also ban gambling-like features such as pay-to-win for minors, require influencers to label ads, and make subscriptions easier to cancel. Companies that break it could be fined up to 6% of worldwide annual turnover. The Commission says it could save EU consumers up to €13.7 billion a year; some gaming groups call it 'unworkable, unjustified and disproportionate'.

Everything else