OpenAI Confidentially Files for IPO, Following Anthropic

OpenAI has filed a confidential S-1 with the SEC, joining Anthropic in the race to go public — a race that, let’s be honest, has less to do with needing cash and more to do with neither company wanting the other to get there first. The filing comes with the standard caveat that timing hasn’t been determined, which is corporate for “we’ll tell you when we’re ready.” What I want to see is the S-1 itself: the section where they explain the business model of spending $7 billion a year to get to AGI should be a real page-turner.

Microsoft Hacked to Deliver Malware to Claude and Gemini Users

This one deserves your full attention: hackers compromised Microsoft’s GitHub repositories — for the second time in weeks, per Ars Technica — and pushed credential-stealing malware that activated the moment an AI coding agent opened the packages. Microsoft responded by nuking over 70 of its own repos. The attack was specifically designed to prey on AI agents doing automated tasks, which is a preview of the security landscape we’re walking into. OpenAI launched “Lockdown Mode” last week to protect against prompt injection; turns out the supply chain is just as juicy a target.

Meta Deletes Face-Recognition System From Its Smart Glasses App After WIRED Report

WIRED found facial recognition code baked into the Meta AI companion app for Ray-Ban smart glasses, published the story, and — poof — Meta quietly yanked the code from the latest version without a word of explanation. No statement. No acknowledgment. Just a silent deletion and a company “declining to comment.” I’ll give Meta credit for moving fast, but the part where they won’t say whether it’s coming back permanently is doing a lot of heavy lifting here.

Google Just Redesigned the Search Box for the First Time in 25 Years

The thin white rectangle with a blinking cursor — arguably the most-used interface in the history of computing — is getting its first overhaul since the Clinton administration. Google’s redesign at I/O isn’t cosmetic; it’s a structural rethinking of what a search query even is in an era where AI can handle multi-part, conversational requests. When the input box changes, the entire assumption about how humans interact with information changes with it. Twenty-five years is a long run for a text field.

NotebookLM’s Gemini 3.5 Upgrade Adds a Cloud Computer and Help Finding Sources

Google is rolling out a meaningful upgrade to NotebookLM, swapping in the Gemini 3.5 model and adding a cloud computer capability — meaning it can now go out and actually do things, not just summarize what you hand it. Source-finding assistance is a genuinely useful addition for anyone using it for research. NotebookLM has quietly become one of the most practically useful AI tools out there, and Google seems to know it; this is the kind of steady, unglamorous product work that compounds over time.

Apple’s WWDC AI Demos Looked More Real After $250M False Ad Settlement

There’s a reason this year’s WWDC keynote felt different — Apple paid $250 million to settle a false advertising lawsuit over AI feature demos that never shipped, and apparently learned a lesson about showing things that actually work. The demos this year featured real people holding real phones doing real things, which, in the context of Apple’s recent history, is almost a radical act of honesty. Nothing like a nine-figure legal bill to sharpen your commitment to shipping what you demo.

The UK Is Betting on a Billion-Dollar AI Supercomputer to Kick Its Addiction to US Tech

Britain is pushing a state-backed AI supercomputer initiative specifically designed to reduce dependence on American cloud infrastructure and supercharge homegrown chip startups. It’s a geopolitically savvy move — Europe (and the UK post-Brexit) has watched the US consolidate AI infrastructure dominance and decided that’s an uncomfortable place to be in a world where AI increasingly underpins national security. Whether a government-led initiative can actually compete with the capital and speed of Nvidia, AWS, and Azure is the billion-dollar question. Emphasis on billion.

Momfluencers Are Pitching AI as a Better ‘Coparent’ Than Men

A growing cohort of “momfluencers” is selling courses on using ChatGPT to handle the mental load of household management — meal planning, scheduling, research, the endless cognitive overhead of running a family — and framing it explicitly as a better partner than their actual partners. It’s funny until it isn’t: this is a real behavioral shift with real implications for how AI embeds itself into domestic life, and the fact that it’s overwhelmingly women adopting these tools for unpaid labor says something uncomfortable about whose time we’ve decided is worth automating.

Bottom Line

The week’s throughline is trust: trust in AI agents not to run malware, trust in Apple to ship what it demos, trust in Meta not to build a face recognition database into your sunglasses — and the market, apparently, trusting both OpenAI and Anthropic enough to take their money.