AI This Week

French lab Mistral AI has raised $830 million in debt to build a new data center near Paris that will be powered by Nvidia chips. The money fuels a massive infrastructure push. The data center will be powered by 13,800 Nvidia GB300 graphics processing units (GPUs), bringing its total capacity to 44 MW. It plans to complete building the data center in Bruyères-le-Châtel and make it operational in the second quarter of 2026. Mistral aims to have 200 MW of capacity across Europe by the end of 2027. The push is about sovereignty. "Scaling our infrastructure in Europe is critical to empower our customers and to ensure AI innovation and autonomy remain at the heart of Europe," CEO Arthur Mensch declared.

Anthropic accidentally exposed approximately 500,000 lines of source code across nearly 2,000 files when a file used internally for debugging was accidentally bundled into a routine update of Claude Code and pushed to the public registry. The npm package containing Claude Code v2.1.88 exposed the source map file, which contained the full and unobfuscated TypeScript source code. The leak potentially allows a competitor to reverse-engineer how Claude Code's agentic harness works and use that knowledge to improve their own products. The leak marks Anthropic's second major data blunder in under a week, following the discovery of descriptions of the company's upcoming AI model in a publicly accessible data cache. Anthropic confirmed this was a release packaging issue caused by human error, not a security breach.

‍

OpenAI closed its record-breaking funding round at a post-money valuation of $852 billion, with $122 billion in committed capital, up from the previously announced $110 billion figure. SoftBank co-led the round alongside investors including Andreessen Horowitz and D.E. Shaw Ventures. ChatGPT now supports more than 900 million weekly active users, including more than 50 million subscribers. OpenAI extended participation to investors through bank channels for the first time and raised $3 billion from individual investors. CEO Sam Altman will face pressure to justify the massive valuation as the company gears up for a potential IPO. The startup has been retreating from hefty spending plans and shuttering features including its short-form video app Sora to rein in costs.

‍

Microsoft just made its AI assistants collaborate. The company announced a new "Critique" feature for its Copilot Researcher tool that combines OpenAI's ChatGPT and Anthropic's Claude in a single workflow. Here's how it works: GPT generates initial responses. Then Claude refines them. Microsoft says this feedback loop delivers stronger results for factual accuracy and analysis. The company also introduced "Model Council," which takes a different approach. It shows side-by-side responses from both AI models and generates a report highlighting where they agree and disagree. Users can choose either method. Both features are now available through Microsoft 365 Copilot's Frontier early access program. Microsoft claims the upgrade outperforms Perplexity's Deep Research models on accuracy and completeness benchmarks.

‍

‍

A data mishap has exposed Anthropic's closely guarded secrets. Nearly 3,000 internal assets spilled onto the public web after the company failed to mark files as private in its content management system.

The leaked materials reveal "Claude Mythos," which Anthropic calls "by far the most powerful AI model we've ever developed." The company has confirmed the leak and says the model represents a "step change" in AI performance. It's currently in trials with select early access customers.

The leak also unveiled a new unreleased model tier called "Capybara," designed to sit above Anthropic's existing Opus tier. Most striking: internal documents warn that Claude Mythos poses serious cybersecurity risks and could help hackers run cyberattacks that "farout pace the efforts of defenders."

‍

Apple’s iOS 27 update will reportedly let users choose which AI chatbot connects to Siri, expanding beyond the current ChatGPT tie-in. A report citing Bloomberg’s Mark Gurman says third-party assistants downloaded from the App Store — including Google’s Gemini and Anthropic’s Claude — could supply answers through Siri on iPhone, iPad, and Mac. The system, said to be called “Extensions,” would let users turn specific chatbots on or off. The same integrations may also support a standalone app for an AI-upgraded Siri that can take actions across apps on a user’s behalf. After setbacks, Apple said in January it is working with Google on its Siri overhaul.

Reddit is rolling out new “human verification” checks aimed at accounts that look automated, while also labeling service-style automated accounts with an “APP” tag. It won’t apply sitewide. The company says verification will trigger only when activity patterns or technical signals suggest a bot, such as unusually fast posting. Accounts that fail may face restrictions. Reddit plans to use third-party options including passkeys from Apple, Google, and YubiKey, biometric tools like Face ID, and World ID, and in some regions government IDs due to age-verification rules. CEO Steve Huffman said Reddit wants proof of a person, not their identity, to keep anonymity intact. Reddit also removes about 100,000 bot or spam accounts daily.

OpenAI will shut down its Sora AI video app just months after launch, telling users it will share timelines for the app and API and explain how creators can preserve their work. The move also unwinds a major Hollywood tie-up. A source familiar with the matter says Disney is exiting its deal with OpenAI, a pact announced last year that included a planned $1 billion investment and licensing of Disney characters for Sora, with an eye toward eventual Disney+ integration. Sora’s debut last fall stunned the industry, then prompted a quick reset to give studios and talent more control over IP and likenesses. OpenAI says it will keep working on AI video within other products.

Anthropic has announced that its Claude Code and Claude Cowork tools are being updated to accomplish tasks using your computer, with these AI resources becoming capableof opening files, using the browser and running dev tools. When enabled, the Claude AI chatbot will first prioritize connectors to supported services suchas the Google workplace suite or Slack, but if a connector isn't available, it will be able to still execute an assigned task. Claude should ask for permission before taking these actions, but Anthropic still recommends not using this feature to handle sensitive information as a precaution. Claude computer use will initially be available to Claude Pro and Claude Max subscribers on macOS, with this feature still in a research preview that will continue to be adjusted based on Anthropic's user feedback.

Elon Musk has launched TeraFab, a $25 billion joint chip fabrication venture between Tesla, SpaceX, and xAI targeting one terawatt of AI compute annually. Taking the stage at Austin's defunct Seaholm Power Plant, Musk called the project "the most epic chip building exercise in history by far." The facility will consolidate every stage of semiconductor production under one roof, including chip design, lithography, fabrication, memory production, advanced packaging, and testing. The chips are expected to be built on a 2nm process, one of the most advanced manufacturing technologies. TeraFab will produce inference chips for Tesla vehicles and Optimus robots, plus D3 chips custom-designed for orbital AI satellites. Musk said 80% of TeraFab's compute output would be directed toward space-based orbital AI satellites. Neither Tesla nor SpaceX has prior experience operating a semiconductor fabrication facility.