Topics
Artificial intelligence and the technology around it: models, agents, tools and the companies behind them, explained in depth.
Between June and August, Z.ai, Moonshot AI and Alibaba put the weights of their flagship models online: GLM-5.2 under MIT, Kimi K3 at 2.8 trillion parameters, Qwen3.8 in two sizes. The fine print differs sharply, and for large companies it now matters more than the parameter count.
A June executive order lets US national security agencies test the most capable AI models for up to 30 days before release, on a voluntary basis. Ten days later, export controls took Anthropic's Claude Fable 5 and Mythos 5 offline worldwide. What each side said, how access came back, and what it means for users outside the United States.
Since 2 August 2026, chatbots in the EU must say they are AI and generated media must carry machine readable marks. The Digital Omnibus pushed the heavier high risk rules to December 2027 and August 2028 and added a ban on nudifier apps. What applies now, and a checklist.
Spotify is rolling out an "AI Persona" badge that keeps artists with AI generated identities out of its recommendations unless listeners follow them. Deezer says fully AI generated tracks now make up more than half of its daily uploads, but only 1 to 3 percent of streams.
OpenAI's GPT-6 Astra is the first model the company rates Critical for cybersecurity, so its most offensive skills sit behind a vetted access programme. Anthropic shipped two frontier models in the same month. The pattern: stronger models, narrower default access and cheaper running costs.
We added 16 reports on what happened in AI between April and September 2026. Each one is dated to the day of the event and says at the top when we added it; search engines get the same two dates. Here is why, the full list, and what changed in older stories.
Washington now says "Super Intelligence" where it used to say artificial intelligence. The abbreviation happens to be Slovenia's internet country code, and the national registry now logs more new .si domains in a day than it used to see in a month.
Micron reported record quarterly revenue of USD 54.2 billion on 30 September 2026, up 379 percent in a year, at a gross margin of 86.8 percent. It expects memory to be even scarcer in 2027 and 2028, which keeps pressure on the price of servers, phones and laptops.
Google DeepMind's first new flagship generation since Gemini 3 raises the output limit to 1 million tokens and launches at USD 2 per million input tokens. On 30 September only a small group of security teams could use it, and Google gave no date for everyone else.
ElevenLabs released Eleven v4 and Eleven v4 Turbo on 28 September 2026: a more expressive speech model directed with plain language tags, and a low latency version for voice agents. Both cover more than 90 languages and sell at a 72 percent API discount until 12 October.
On 24 September 2026 OpenAI removed the Videos API and every Sora 2 model from its platform, finishing a shutdown it announced in March. Developers got no replacement, and app users had until April to export their videos before deletion.
Anthropic released Claude Opus 5.5 on 22 September 2026 at USD 4 and USD 20 per million tokens, saying it matches Claude Fable 5.1 on most work. It also inherits Fable style safeguards: most cybersecurity tasks go to Opus 4.8, and biology research blocked by the safeguards needs a verification programme.