Here's a number worth sitting with: OpenAI, the company that put AI on everyone's lips, is currently valued at less than its biggest rival. Anthropic just crossed a $965 billion post-money valuation after its latest fundraise, edging past OpenAI's $852 billion. And now both companies are sprinting toward the same finish line — a public stock market listing.

OpenAI confirmed Monday that it has confidentially submitted a Form S-1 to the SEC, following Anthropic's identical move earlier this month. A confidential filing is essentially a soft launch for an IPO — the paperwork is in, but the juicy details like executive pay, risk factors, and full financials stay locked away until the company decides to go fully public with the document. It's a way to test the regulatory waters without fully exposing yourself to public scrutiny.

The timing matters a lot here. This isn't just two AI companies doing routine corporate housekeeping. It's a legitimately competitive race, and the outcome will shape how Wall Street values the entire AI sector for years. Whichever company debuts first — and at a stronger valuation — gets to set the narrative.

But OpenAI's road to the public markets hasn't been smooth. Reports have surfaced that CFO Sarah Friar has been notably less enthusiastic about the accelerated timeline than CEO Sam Altman. The concerns aren't trivial: missed revenue targets, slower-than-expected user growth, and a compute spending commitment that raises eyebrows. OpenAI originally floated plans to spend $1.4 trillion on compute infrastructure — a figure so large that Altman reportedly got defensive when pressed on it publicly. The company later revised that number down to $600 billion through 2030, which is still an almost incomprehensible amount of money to commit before you've ever sold a share on the open market.

Then there's the SpaceX factor. Elon Musk's rocket company is targeting a June 12 IPO that could raise $80 billion and potentially become the largest public debut in history. OpenAI's listing will inevitably be measured against that spectacle — especially given the tangled web connecting these companies. SpaceX acquired OpenAI competitor xAI, and separately inked a deal with Anthropic worth $15 billion a year for data center access. Meanwhile, the legal dust from the Musk v. Altman trial has barely settled.

What makes this moment genuinely fascinating is that we're watching the AI industry grow up in real time. A year ago, these were private companies burning venture capital and making grand promises. Now they're filing paperwork with the SEC, hiring investor relations teams, and preparing to answer to shareholders every quarter. The freewheeling startup era of AI is ending. The era of quarterly earnings calls is beginning.

Whether investors will treat OpenAI like a transformational technology platform or an extremely expensive bet on future revenue remains the central question. The S-1, when it goes public, will be one of the most-read documents in recent Wall Street history.

Here is the part that should stop you mid-scroll: a Fields Medal winner — the mathematician's equivalent of a Nobel Prize — called an AI's solution to an 80-year-old problem a genuine milestone. Not a party trick. Not a benchmark flex. A milestone.

The problem in question is the Erdős unit distance conjecture, a deceptively simple geometry puzzle first posed in 1946. Picture a bunch of dots scattered on a flat plane. Now count how many pairs of dots sit exactly one unit of distance apart. The question Paul Erdős asked was: how many such pairs can you pack into a set of points before the geometry simply won't allow any more? Mathematicians had been wrestling with the upper bound of that answer for eight decades. OpenAI's model didn't just make progress on it — it disproved the conjecture entirely.

Fields Medalist Tim Gowers publicly called it a milestone in AI mathematics. University of Toronto professor Daniel Litt went a step further, saying it was the first AI result he found exciting on its own terms, not just as a sign of things to come. That is a meaningful distinction. Mathematicians are a skeptical crowd.

So how did the model actually do it? Not through some alien form of reasoning we've never seen before. The AI drew on existing techniques spread across several subfields of mathematics and stitched them together into a complete proof. It was more like a very well-read, extremely tireless research assistant than a genius having a eureka moment in the shower. The proof has since been cleaned up and extended by human mathematicians, which tells you something important about where we actually are.

This is the latest step in a progression that has moved faster than almost anyone predicted. Three years ago, large language models were fumbling basic arithmetic. Last year, they started acing high school math competitions. Earlier this year, they were contributing to research in constrained settings, with heavy human interpretation required to turn their outputs into anything publishable. Now one has independently resolved a major open conjecture.

The most honest framing right now is that AI and human mathematicians are better together than apart. AI systems have effectively read everything — every paper, every proof, every clever trick from every subfield — and they will grind through hundreds of dead-end proof strategies without complaint. Humans, meanwhile, can still think more deeply about a single hard problem and ask the kinds of questions that point research in genuinely new directions.

But that division of labor may not hold. AI capabilities in mathematics have been compounding so quickly that serious people are now openly asking what role human mathematicians will play ten years from now. That is not a dismissive question — it is an honest one.

Paul Erdős, who posed the original problem, wrote more than 1,500 mathematical papers in his lifetime, more than any other mathematician in history. His gift was finding problems that fit in a sentence but contained worlds of hidden complexity. It is a little poetic that a problem he dreamed up became the one that marked a new era in machine intelligence.

Here's a sentence that would have sounded like science fiction three years ago: the AI model Anthropic builds to be safe and helpful is now writing more than 80 percent of the code used to build itself. That's not a demo. That's production.

Anthropic has confirmed that Claude authors the overwhelming majority of new code shipped at the company — not prototypes, not internal experiments, but the real software running in the background of one of the most closely watched AI labs in the world. It's a milestone that reframes the conversation about what AI-assisted development actually looks like at scale.

For most companies still debating whether to let developers use AI coding assistants, Anthropic has lapped the field. This isn't a story about productivity gains measured in percentage points. It's a story about a fundamental shift in who — or what — is doing the work of building software.

The implications for enterprise software teams are significant and a little unsettling depending on where you sit. If a frontier AI lab is running at 80-plus percent AI-authored code, the productivity gap between organizations that have fully embraced these tools and those still treating them as optional add-ons is only going to widen. Fast.

It's worth pausing on what "authored by Claude" actually means in practice. This isn't Claude operating autonomously in a vacuum. Engineers at Anthropic are still directing the work — setting goals, reviewing outputs, catching mistakes, and making architectural decisions. But the ratio of human keystrokes to AI-generated code has inverted dramatically. The human is increasingly the editor, not the author.

That shift carries real consequences for how software teams need to be structured and what skills matter most. Writing code fluently is becoming less valuable than being able to evaluate code quickly, catch subtle errors, and think clearly about system design at a higher level. The bottleneck is moving from implementation to judgment.

For enterprises trying to figure out how to keep up, the Anthropic example is instructive even if it's extreme. Most large organizations are nowhere near 80 percent. They're wrestling with questions about security, IP ownership, compliance, and how to train existing engineers to work effectively alongside AI tools. Those are real and legitimate concerns — but they can't be used indefinitely as reasons to move slowly while competitors accelerate.

The other thing worth noting is what this signal means for Anthropic's own roadmap. A company that is using its own model to write most of its code has enormous incentive to keep improving that model's coding ability. Every gain Claude makes in software development makes Anthropic's own engineering operation faster. The feedback loop here is tight and self-reinforcing.

We are, in short, watching AI development eat its own tail in the best possible way — and the pace of that loop is only going to pick up.

Meta was shipping face-recognition software to millions of phones while publicly telling the world it was still "thinking through" whether to build the thing at all. That gap between what the company said and what it actually did is the story here — and it's a significant one.

WIRED's analysis of Meta's AI app uncovered a dormant feature called NameTag buried inside software that has been downloaded more than 50 million times. The app is required to run key functions on Meta's Ray-Ban and Oakley smart glasses. NameTag, as the code reveals, is designed to identify faces captured through the glasses' camera, convert them into unique biometric signatures, and alert the wearer when it recognizes someone nearby.

The feature isn't live yet. But that's almost beside the point. Core components — three separate AI models that detect faces, crop them, and encode them into biometric data — have already been pulled from Meta's servers and are sitting on users' phones right now. The infrastructure is in place. The switch just hasn't been flipped.

Meta's public posture on this makes the discovery more uncomfortable. In April, a company spokesperson said that if face recognition were ever deployed, it would be done with "a very thoughtful approach." WIRED found that as early as January, the underlying system was already being quietly integrated into updates pushed to millions of devices. That's not thoughtful consideration — that's a feature in staging.

This isn't Meta's first time in this particular courtroom. The company shut down its Facebook photo-tagging face recognition system in 2021, deleting over a billion stored faceprints after years of backlash. That decision came after Meta paid $650 million to settle a class-action lawsuit from Illinois users, and the legal trouble didn't stop there — in 2024, the company agreed to a $1.4 billion settlement with Texas over separate allegations of unlawful biometric data collection.

So the company knows exactly what this technology costs when things go wrong. Which raises the obvious question: why build it again, quietly, inside a glasses app?

The timing adds another layer of concern. Internal documents reported by The New York Times in February showed Meta had considered launching the feature during a period it described as a "dynamic political environment" — essentially a window when its loudest critics might be distracted by other things. That's not a privacy-first mindset. That's a launch strategy.

For everyday users, the implications are hard to overstate. Smart glasses are nearly invisible as a recording device. Pair them with real-time face recognition and you have a tool that can identify strangers on the street, in coffee shops, at protests. Privacy advocates have warned for years that consumer-grade facial recognition hands dangerous capability to anyone who wants it — stalkers, bad actors, or anyone with a grudge and a pair of fashionable frames.

A later version of the app rebrands NameTag as "Connections," with friendly copy inviting users to "remember the people you met." The reframe is smooth. The underlying technology is the same.