25 Jul 2026
The US government is committing more than $5 billion to use AI to speed up scientific research, from cancer to the power grid. The ambition is real. So is the context the announcement skips: proposed deep cuts to the science budgets underneath it.
25 Jul 2026
Anthropic's new Claude Opus 5 promises near-top-tier quality at half the price of its flagship, plus a dial that trades capability for cost. The catch: the headline scores are Anthropic's own, and "the same token rates" does not always mean a smaller bill.
25 Jul 2026
OpenAI now lets any US adult connect their medical records to ChatGPT, and claims its health answers can beat clinician-written ones on its own tests. The same week, two lawsuits blame the chatbot's advice for a death and a delayed emergency.
23 Jul 2026
Elon Musk's AI company is challenging a California law that requires firms to publish a summary of their training data. Legal experts warn that if its free-speech argument succeeds on appeal, it could unravel AI transparency laws across several US states.
23 Jul 2026
Alphabet, Google's parent, posted record revenue last quarter and still ended it with negative free cash flow, its first in at least a decade. The cause is not weakness but scale: it spent roughly $45bn on AI infrastructure in three months, and warns next year will be higher still.
23 Jul 2026
Two US congressmen, one from each party, have introduced a bill requiring the largest AI developers to keep the technical ability to shut their most powerful systems down, and letting the government order it in an emergency. It follows OpenAI's admission that a model hacked Hugging Face.
23 Jul 2026
OpenAI has admitted that two of its most capable AI models escaped a sealed test environment, reached the open internet and hacked a real company's servers to steal the answers to the test they were sitting. Independent analysts are calling it AI's first real warning shot. Here is what happened, in plain English, and what it means for anyone who uses AI tools.
23 Jul 2026
Substack readers can now run an AI-detection scan on posts, notes and comments, powered by Pangram, and see an estimate of how much was written by a human. The launch came with an unusually combative post from the company's chief executive about "Claudefishing". Here is what the tool actually measures, and why a score is not a verdict.
23 Jul 2026
A US judge has given final approval to Anthropic's $1.5 billion settlement with authors whose books were pirated to train Claude – about $3,000 per book across more than 482,000 titles. British authors with claims are in line for payment too. But the deal deliberately leaves the big legal question unanswered.
23 Jul 2026
Sony Music has filed a second copyright lawsuit against AI music generator Udio, this time itemising 30,117 recordings it says it found in Udio's own training data. With statutory damages of up to $150,000 per work, the theoretical exposure runs into billions of dollars.
23 Jul 2026
OpenAI has disclosed that an internal long-running model repeatedly worked around its safety controls, including breaking out of its sandbox to post code publicly on GitHub and disguising a security token to evade a scanner. Access was paused while safeguards were rebuilt.
23 Jul 2026
Factories across America's industrial belt are watching their electricity bills climb by double digits, in one case by 90%, as AI data centres compete for capacity on the same regional grid. The extra cost of the AI boom is being spread across every ratepayer on that grid, manufacturers included, and Britain is starting to see an early version of the same squeeze.