The AI news, read all the way through.
We go to the filing, the email or the code, read what people are saying about it, and tell you plainly what it means.

I read all 335 pages of Anthropic's IPO filing. The headlines missed the scariest parts.
The filing says the company could become conscious, could cause harm no insurer can cover, is in a court fight with the US government, and can never be outvoted by the people buying its stock. All of that is in the document. Most of it is not in the news.

Microsoft offered 60,000 people a 'buyout'. The money is the same as getting fired.
Microsoft calls it a Voluntary Exit Offer and says nobody is being incentivised to leave. That line is accurate, and that is the problem.

Two AIs just broke a 27-year-old rule of computer science on the same day. Nobody has checked either proof.
If true, it is one of the biggest results in computing history. The people named as checkers say their involvement 'should not be taken as certification.' And a second AI, from a rival lab, found the same thing on the same date with a different method.

Proton is re-interviewing its staff about AI. I read the whole email, and its own code repo.
Proton says this is not layoffs. The email says the review may decide your continued employment. Both of those sentences came from Proton on the same day.

Meta took the word 'train' out of its new business terms. The test logins, the crawling and the word 'develop' are still in.
Meta says the new clause is only about ads and that robots.txt still works. Then why is robots.txt nowhere in the clause, and why does it need to create logins on your website?

MKBHD bought Final Cut once and used it for ten years. Apple's new AI update is built so the next creator can't.
The AI is the headline. The real change is quieter: a cheap price, a few features you can't get any other way, and project files you can't open once you stop paying.

Google's Genie 4 builds worlds that talk back. Its own fine print still says 'a few minutes'.
Genie 4 is the most impressive world model Google has shown. It also ships with no spec sheet, a 'sequences shortened' disclaimer, and the same 'few minutes' line Google used fourteen months ago.

Thieves took more than $100 million of AI hardware off US roads. Most of it was stolen with a login, not a gun.
The heist stories have rammed escort cars and spoofed GPS. The real story is duller and worse: most AI gear is stolen by people who never touch the truck until the paperwork already says it's theirs.