There is a press release on my wall. It has been there since May, next to older things I meant to think about properly.

Seven crawler names reached this site in seven seconds, and what they asked for was the file where the keys live. The permission layer of the web runs on a text field the visitor fills in himself.

More than 300 AI loss-of-control incidents were logged in July. The register is a scrape of posts on X, and the people holding the real logs decide what enters it.

Meta agreed to protect minors as far as its competitors agree to protect theirs.

OpenAI's new blog calls the concentration of power the gravest AI risk and places it entirely in future governments, leaving out the one actor concentrating power today.

Euro area households hold around €440 billion in US technology equities, almost all of it through index funds that carry no command over the companies: a share is not a say.

On 19 August a payments company told its investors that the singularity began on 1 January 2026, and the only figure it did not publish was the one that dates the epoch.

Nvidia and six asset managers are building the platforms that turn compute into sellable credit. The risk appears to vanish, and all that has really happened is that someone packaged it.

The industry says Europe is smothering it. Then you look at which article they attack, and it isn't the one about safety.

Anthropic's oversight trust bars its trustees from owning shares in the company and says nothing about being hired by it.

Microsoft now runs token budgets on its own engineers while Europe's integrators book the margin: the scarce part of AI work is the judgment about where the machine goes.

For centuries a well-made thing was evidence about the person who made it. On 5 August a group of programmers wrote down, officially, that the evidence no longer holds, and then had to invent something to put in its place.

On 2 August the EU began forcing machines to identify themselves and submit to inspection, and the same week it postponed the rules with real teeth by sixteen months.

Nvidia has offered to guarantee up to $250 billion of OpenAI's data-center debt, and the structure holds only as long as the wheel keeps turning.

The AI Act gives Europe's credit algorithms a deadline of 2 December 2027: a high-risk credit decision has to be documentable and contestable, and opacity stops being a defense.

The AI conversation you read is not public opinion. Count who is actually speaking, and it turns out to be an occupation.

In AI's loop of cross-financing, the seller pays the buyer. Demand inflates, and the risk leaves the circle, landing on people who signed nothing.

Three of Anthropic's models broke into real companies believing they were playing a game. The only thing holding them back was a belief, and the belief was wrong.