AI News15 Sept 2026
Nvidia and Palantir hold back on Anthropic's models over data storage
Some of the heaviest AI users in the world have stopped taking their suppliers' word on data. According to The Information, a tech news site, Palantir wants Anthropic to guarantee it keeps none of Palantir's data. Nvidia keeps Claude, Anthropic's model, away from sensitive work, and Booz Allen, a US defence contractor, has banned it from cybersecurity projects. The trigger was a June change letting Anthropic store usage logs for 30 days to catch misuse. The same firms are limiting OpenAI's models too. Ask your IT team where your prompts end up, before a customer asks you.
AI News15 Sept 2026
The EU wants AI chatbots age-checked like social media
Europe is about to treat chatbots like social media. On Thursday the European Commission will propose a Kids Act that would keep children under 15 away from social networks, online games and AI chatbots, Reuters and Bloomberg report. Services would have to check ages and give parents control, and the European Parliament and member states still get their say. On Monday, the Confederation of Danish Industry proposed AI tools for Danish pupils from the age of ten. Those two ideas fit together only if somebody builds AI designed for children, and I would love to see a Danish company build it.
AI News15 Sept 2026
Chip stocks fall on the slowdown talk, and Trump calls AI fears a hoax
The debate about slowing AI down reached the stock market on Monday. The main US chip index closed 5.9% lower. SoftBank, one of OpenAI's largest owners, lost almost 11% in Tokyo. That afternoon Donald Trump phoned Jensen Huang, Nvidia's CEO, while Huang was on stage at an investor conference, and called AI fears a hoax. Huang agreed. US government bond yields touched 5% the same day, so AI takes only part of the blame. Investors at least seem to take the warnings more seriously than the President does, and I think they are right to.
AI News15 Sept 2026
China's spy chief warns about AI as Beijing rejects Amodei's slowdown call
Dario Amodei, Anthropic's CEO, called for AI to slow down at the weekend. In the same essay he argued for keeping advanced chips out of China, and Beijing replied within days. China's foreign ministry called it fearmongering that disrupts global cooperation on AI rules. A day earlier Chen Yixin, the minister in charge of China's intelligence service, warned that AI threatens the country's political stability and infrastructure. He singled out the hacking abilities of American models. So Beijing fears the very machines it accuses Amodei of hyping. Trump and Xi Jinping meet on 24 September, and I would not bet on a pause.
AI News15 Sept 2026
Contractors are reading real ChatGPT conversations under Project Lily
OpenAI pays hundreds of contractors to read real ChatGPT conversations and rate the answers. The effort is called Project Lily, according to leaked documents seen by 404 Media, an investigative tech site. Names are removed, but OpenAI's own filter lets some personal details through. Reviewers can also see summaries of what a user has discussed before. Company accounts are excluded by default, while personal accounts are included until you switch it off in settings. Anthropic says it uses human reviewers too. It shows why companies like Palantir, in today's first story, want those logs kept in their own hands.