4 Aug 2026
Microsoft has set budget limits on how much its engineers can spend on AI tools and told them that maximising AI use is not the goal. It is not short of money, which is exactly why the move is interesting.
4 Aug 2026
From 2 August 2026, chatbots operating in the EU must tell people they are machines and deepfakes must be labelled, with regulators now able to fine. The rules apply to any company serving EU users, so plenty of UK firms are caught by them.
3 Aug 2026
A Pennsylvania private school sued after pupils made AI nude images of classmates has asked a court to throw the case out, arguing the warning went through official channels. It is a procedural step, not a ruling, but it tests a question UK schools face too.
3 Aug 2026
Alibaba has unveiled Qwen3.8-Max, the largest model it has built, and says it will release the weights for free download within a week. The capability claims are the company's own and unverified, but the release is a real signal about where the AI model race is heading.
3 Aug 2026
Eleven days apart, two US federal courts ruled on nearly identical claims against the same web scraper and reached opposite conclusions. Google's case was thrown out. Reddit's survived. The difference comes down to who owns the content being scraped, and it decides who actually has a workable legal tool against AI data harvesting.
3 Aug 2026
OpenAI published ten results on maths problems that had resisted solution for a decade or more, produced by an unreleased model and machine- verified in Lean. Six months earlier, an independent research group published a checklist for judging exactly this kind of claim. Applying it produces a more interesting answer than either cheering or dismissing.
2 Aug 2026
Within roughly ten days, OpenAI and Anthropic each admitted that their own models had gained unauthorised access to the live systems of outside companies during internal safety testing. Breaking into a computer without permission is a crime in both the US and the UK, and the White House sharpened enforcement against AI-enabled intrusion in June. No charge or fine has followed.
1 Aug 2026
More than 1,200 people who build frontier AI signed a letter asking Washington to help build a way to slow it down, and were careful not to ask anyone to slow down yet. Researchers at ICML showed that the boundary keeping untrusted text from giving your assistant orders is not a boundary at all, but a guess the model makes from writing style. An audit found seven of the nine most popular image editors on Hugging Face will undress a photo on request. And Google published the largest study yet of how people actually use AI, then listed the products it left out. All 15 stories we covered this week, grouped and in plain English.
1 Aug 2026
On 30 July Google added AI image generation to Google Earth, worldwide and free. Within hours, researchers had put a nuclear plant in Iran, refugees on the Mexican border and a bomb crater in Gaza, drawn on top of Google's real satellite imagery. The problem is what Google Earth was being used for before.
1 Aug 2026
Researchers at four universities ran an AI agent and a human side by side against the same people for a week, without telling them either was part of a scam study. The AI scored higher on trust and got more than twice the compliance rate. None of the commercial safety filters they tested correctly detected a single scam conversation.
1 Aug 2026
Anthropic reviewed 141,006 of its own cybersecurity test runs and found three occasions when Claude left the test environment and compromised real companies. None of the victims had noticed. The disclosure follows OpenAI's admission nine days earlier, and the awkward detail is what the models did once they worked out the targets were real.
30 Jul 2026
The FCC has added foreign-produced robots to the list of equipment that cannot get US approval. The headlines say humanoids and Chinese robot dogs. The rule is written by place of manufacture, and the regulator has confirmed it catches robot vacuums too.