The Week in AI: Oct 26–30, 2026

Anthropic’s IPO filing puts AI’s funding boom to the test

Anthropic has filed to list on Nasdaq under the ticker ANT. There’s no share count or price range yet, but reports put the offering around $100 billion and the valuation somewhere between $1.5 and $2 trillion. That makes it the biggest test yet of how badly investors want AI.

The numbers are bonkers in both directions. Anthropic brought in $16.3 billion in first-half revenue, up nearly 14x year over year, alongside an $80.7 billion net loss. (Most of that loss is a noncash charge tied to convertible notes, but still.) So the real question is whether growth like that can cover the compute bills coming due.

A strong debut could open the door for rivals to raise money; a weak one deepens worries as financing tightens. And read the fine print: Anthropic itself calls enterprise adoption “early,” and its $40 trillion opportunity is the value of all knowledge work, not projected sales.

More coverage:

• NY Times: Anthropic’s IPO S-1 Spells Out a Future of Promise and Peril
• NY Times: In Anthropic I.P.O., Goldman and Morgan Stanley Slug It Out for Top Bank
• NY Times: Meet Anthropic’s Seven Founders as the A.I. Firm Prepares for an IPO
• WSJ: Anthropic’s ‘Adjusted’ Profit Exposes Wall Street’s Earnings Game
• WSJ: Anthropic’s Big IPO Gamble
• WSJ: Anthropic’s IPO Will Be a National-Security Win for Taiwan
• Transformer: Will Anthropic’s post-IPO governance really protect its safety mission?
• TechCrunch: VCs will make billions on Anthropic’s IPO, but Amazon is set to make much (much) more

Anthropic’s $63 billion chip deal exposes AI financing strain

Big Tech’s earnings show AI demand is real, but the payoff is uneven. Microsoft grew Azure 48% and boosted profit despite $45.2 billion in quarterly capital spending. Alphabet’s cloud revenue jumped 103%, but profit fell and cash flow went negative. And Google admitted AI searches earn less than traditional ones. Not ideal when you’re trying to justify the spend.

Meanwhile, the money keeps flowing, just on tougher terms. Meta wants to raise up to $95 billion and SpaceX announced $40 billion in debt, while Musk warned tighter credit could sink smaller data-center players. Then Anthropic’s $63 billion chip-leasing deal nearly fell apart until Broadcom stepped in and accepted higher rates and lender protections. Lenders are still lending. They’re just asking for more.

More coverage:

• WSJ: AI Is Almost the Easy Part. Everything Else Is Hard
• WSJ: Alphabet Still Has the Ingredients to Win in an AI Pullback Despite Stock Price Hit
• NY Times: Investors Brace for Big Tech’s A.I. Spending Bill

Dartmouth provost resigns after review of undisclosed AI use

Dartmouth Provost Santiago Schnell has resigned after a review found his AI use in academic and opinion writing broke publication rules and college standards. He also didn’t disclose it as required and “was not forthcoming,” according to the review.

It started with scrutiny of his published work and calls for his dismissal from students and faculty who felt administrators weren’t held to the same standard as students. Schnell says he wrote his own drafts and used AI to sharpen arguments and clarity. He also questioned whether AI-detection tools are biased.

Dartmouth isn’t alone in fighting over inconsistent AI rules. If the rules are fuzzy for students, apparently they’re fuzzy at the top, too. History professor Matthew Delmont steps in as interim provost through June 2028.

PHP patches race to contain AI-discovered server flaws

PHP has shipped fixes for two critical vulnerabilities across supported 8.x versions, but site operators are still scrambling to patch or pull exposed sites offline. The flaws let unauthenticated attackers run code on servers, and PHP powers an estimated 70% of websites. Worse, attacks started before patches were out, as leaked details and proof-of-concept code spread.

Here’s the twist: researchers found the bugs with OpenAI models, and the PHP Foundation had to rush its release after disclosure details escaped. That’s AI’s double edge in a nutshell. It helps defenders find holes faster, but it helps attackers turn them into exploits faster, too.

More coverage:

• WSJ: AI Hacks Are Coming Thick and Fast and Overwhelming the Internet
• Transformer: PHP code flaw begins the cyberpocalypse — sort of
• Don’t Worry About the Vase: The Hackening
• Noahpinion: The internet breaks
• Exponential View: The internet’s Wuhan moment
• Ryan Orhan (X): AI-driven attacks and major breaches are escalating faster than organizations can handle
• John Gedmark (X): AI-driven vulnerability cascades put critical infrastructure at risk

AI reshapes college teaching as computer science majors decline

AI is forcing colleges to rethink what students learn and how they’re taught. U.S. computer-science programs lost about 53,000 majors last year. Cybersecurity and data science are growing, but AI is reshaping entry-level coding jobs, and some schools are answering with degrees that pair computing with other fields.

The classroom is still a battleground. Tyler Cowen wants colleges to go all in: heavy AI use paired with rigorous independent writing, student-led experiments, even an AI-run course. Researchers counter that the evidence AI improves learning is thin and want the essential thinking left to students. At Berkeley, math professor Richard Borcherds now sees “no point” in grading homework for correctness. Librarians want a say as AI seeps into research tools. And one modest proposal: automate the administrators first.