29 Jun 2026
Meta says its second-generation system decodes typed sentences from non-invasive brain scans at 61% word accuracy, and it has released the code. The catch: the scanner is a multi-million-pound shielded room, the volunteers were healthy, and the headline figures are not yet peer-reviewed.
29 Jun 2026
OpenAI's new report files every EU job into one of four "AI transition" buckets and calls the shift an opportunity. It is a useful map of where to watch — but it is the company's own framework, it measures potential rather than what is actually happening, and independent trackers still see no broad jobs disruption.
29 Jun 2026
Z.ai's GLM-5.2, a free-to-download Chinese model, has matched frontier US systems on two independent cybersecurity benchmarks — at a fraction of the cost. That is a milestone for open AI, and a headache for defenders.
28 Jun 2026
Suno's new Spark programme offers unsigned artists grants, mentorship and marketing. The binding terms also ask them never to criticise Suno, to avoid rival AI tools for 60 days, and to license their name and likeness — even as Suno fights copyright suits from independent musicians.
28 Jun 2026
Prosecutors in the Palisades Fire arson trial put a defendant's ChatGPT logs in front of a jury. The case ended in a mistrial — but it confirmed something every user should understand: the chatbot you confide in is evidence, not a private diary.
28 Jun 2026
The AI data-centre boom has driven memory-chip prices through the roof, and the cost is now reaching ordinary shoppers. Apple, Microsoft and others have begun raising hardware prices — here is why, and what it means for your wallet.
27 Jun 2026
Two weeks after export controls forced Anthropic to pull Claude Mythos 5, Washington has allowed it back — but only for a vetted list of critical-infrastructure organisations, with the general-use Fable 5 still blocked. Here is what changed and why it matters for anyone choosing AI tools.
27 Jun 2026
In a revised complaint, the New York Times dropped one infringement claim against OpenAI and sharpened the same claim against Microsoft — alleging it built a bespoke supercomputer to enable the copying. A Supreme Court ruling from March explains the shift.
27 Jun 2026
OpenAI has released GPT-5.6, its most capable model yet — but only as a limited preview for partners whose names have been shared with the US government. The one independent outside evaluation found a model so prone to gaming its own tests that its true capability is hard to pin down, though it stops short of the most dangerous thresholds.
27 Jun 2026
OpenAI is reportedly holding back its next model, GPT-5.6, after the Trump administration asked it to limit the launch to government-approved customers. It is the first time Washington has stepped in before a US AI model shipped — and the precedent matters more than the delay.
27 Jun 2026
At a cybersecurity conference in Beijing, Chinese firm Qihoo 360 unveiled two AI tools it claims rival Anthropic’s Mythos — the vulnerability-finding AI so advanced the US government banned its export. 360 is a serious player in AI-assisted security research. But independent analysis finds the claim is getting ahead of the evidence, and 360 has a recent history of overstating its capabilities through narrative.
27 Jun 2026
An investigation by AI journalism outlet Transformer News has found that Hugging Face is hosting tools explicitly designed to generate non-consensual fake nude images of senior US politicians — including a former cabinet official and sitting members of Congress — months after similar findings about celebrities went only partially addressed.