Daily Briefing
24 September 2026 · 5 stories
🔥 Top story
1Trump and Xi treat AI as a balance-of-power issue AND a shared risk — a historic first
For the first time, the two most powerful leaders in the world are treating artificial intelligence not as a product to sell, but as a diplomatic subject in its own right. On Thursday 24 September at the White House, Donald Trump and Xi Jinping put AI on the official agenda of their summit, alongside trade, Iran and Taiwan. It is the second meeting of the year — the first took place in Beijing in May. For the general public, it is comparable to the day two heads of state started talking seriously about the climate: AI is leaving the « technology » aisle and entering geopolitics. Experts agree: no binding agreement on AI safety is expected, but the simple fact of discussing it together, at this level, sets a precedent. A major US think tank is already organising a public « debrief » the next morning. Source : https://www.csis.org/programs/trump-xi-2026-summits
2Two US lawmakers want to ban « super-AI » and create a dedicated cabinet department — 20 years in prison for offenders
On Wednesday 23 September, Senator Bernie Sanders and Representative Greg Casar introduced in the US Congress a bill as radical as it is unprecedented: the « Ban Artificial Superintelligence Act ». Four concrete measures: (1) permanently ban AI systems that would exceed human cognitive abilities, (2) freeze advanced development until a federal regulator is in place, (3) create a full-blown Department of AI at cabinet level, (4) punish offenders with up to 20 years in prison. The bill is endorsed by researchers from OpenAI's safety teams and by a Google DeepMind engineer — not only by politicians. For the general public, imagine a country suddenly deciding to ban cars that go above 250 km/h and creating a « Speed Police »: that is exactly the register, but transposed to AI. The bill has little chance of passing in a Republican-majority Congress, but it sends a maximalist political signal 24 hours before the Trump-Xi summit. Source : https://www.sanders.senate.gov/press-releases/news-sanders-casar-introduce-legislation-to-create-new-federal-agency-to-ban-artificial-superintelligence-pause-advanced-ai-development
3A German open-source AI that teaches robots to work — already running in Audi factories
A small German company, Black Forest Labs (the researchers behind Stable Diffusion), released on Wednesday 23 September a brand-new, fully open AI model specialised in robotics: FLUX 3 Action. Its special feature: it looks at a camera image, the state of a robot arm, and an instruction (« pick up that bolt ») — then it decides the 32 micro-movements to chain together to get there. And before each gesture, it imagines what will happen: if the arm grabs the bolt, where will it fall? On NVIDIA's official benchmark, it scores 42.9%, that is 6.1 points better than NVIDIA's competing version — with half the parameters, so it runs faster. Car maker Audi is already using it on its production lines. For the general public, it is as if an engineering school published for free both its textbooks and its training grounds for robots: anyone can now build their own mechanical assistant. Source : https://bfl.ai/blog/flux-3
4A London startup raises 140 million dollars to make medicines with AI — with NATO as an investor
Basecamp Research, a young company based in London with labs in Cambridge, Massachusetts, closed on Wednesday 23 September a 140-million-dollar funding round, valuing it at 800 million. The round is led by a Google-affiliated fund, and brings together an unusual set of investors: NVIDIA (the chips), a fund of Anthropic (the lab behind the ChatGPT competitor), NATO's innovation fund, and the Rockefeller Foundation. The company uses a massive database of genomes collected in 30 countries to train its own AI models, called EDEN. These models then propose drug candidates, the first of which are entering clinical trials this year. For the general public, imagine a giant cookbook library that has digitised the kitchens of the whole world, and whose best recipes are then tested in real restaurants: that is the same idea, but applied to pharmaceuticals. Source : https://www.prnewswire.com/news-releases/basecamp-research-raises-140m-to-advance-ai-designed-therapeutics-302887617.html
5Cyera, the startup that locks down the data AI agents access, raises 400 million — now worth more than 12 billion
Cyera, a company specialising in data security for artificial intelligences, announced on Tuesday 22 September a 400-million-dollar extension to its last funding round, led by Goldman Sachs. Its valuation now exceeds 12 billion. Why such a rush? Because companies have given their AI agents access to their Salesforce files, their Databricks databases, their Slack messages, their HR tools — and no one today knows exactly what those agents do in there. Cyera installs a « data guard » that traces every access and blocks whatever falls outside the authorised scope. More than 100 customers already pay over 100,000 dollars a year for this service, and 14 of them exceed one million. For the general public, it is as if a company posted a guard at the entrance of every room a digital assistant can enter: he logs who goes where, and blocks the exit if someone tries to take an unauthorised document out. Source : https://siliconangle.com/2026/09/22/cyera-raises-another-400m-amid-ai-agent-security-push/
📡 To watch
Chinese lab DeepSeek said to be training its models on Chinese chips rather than American ones
According to a South Korean report dated 22 September, DeepSeek — the Chinese lab whose R1 model shook the world in January 2025 — is preparing to shift its training phase onto Huawei Ascend chips, Chinese processors designed for AI. Goal: become independent of US supplier NVIDIA, whose sales to China have been restricted by the United States for several years. For the general public, it is as if a European carmaker, cut off from German engines, started making its own: the constraint speeds up re-industrialisation instead of slowing it down. Source : https://www.chosun.com/english/industry-en/2026/09/22/3JM3NLWBWBDVZN3Z5T2L45U2ZI/
Alibaba prepares Qwen 4 and wants to become the operating system of Chinese smartphones
Chinese cloud giant Alibaba confirmed on Tuesday 22 September that the new version of its AI model, Qwen 4, has entered its training phase. Its « Qwen Intelligence » assistant will be embedded from 28 September in the next Honor Magic9 phone, with a 91.8% success rate on common tasks. Alibaba is aiming for the place of « operating system for AI » across all Chinese devices. For the general public, it is as if a software editor jumped from « application provider » to « phone manufacturer ». Source : https://www.thestar.com.my/tech/tech-news/alibaba-eyes-revenue-share-for-open-source-ai-system
AI Engineer Paris opens its doors at Station F — 1,000 AI engineers meet while Trump and Xi talk in Washington
The second Paris edition of the AI Engineer conference opens on Wednesday evening 23 September and runs all day Thursday 24 September at Station F, Paris 13ᵉ. On the programme: 1,000 AI engineers, 30 talks, 16 workshops, 24 stands — with Mistral AI (the French lab) as main organiser. While the American and Chinese heads of state negotiate in Washington, Paris hosts those who build the tools day-to-day. For the general public, it is the European version of a big trade show: no diplomatic spotlight, but the sound of keyboards. Source : https://ai.engineer/paris/2026
Joint announcements expected at the end of the Trump-Xi summit this Thursday evening
The Trump-Xi summit of Thursday 24 September is to conclude with a joint press conference late in the evening (between 22:00 and midnight, Paris time). Three threads are being watched closely: (1) a possible joint statement on AI, even non-binding, (2) the US position on high-end NVIDIA chip sales to China, (3) a bilateral working frame on autonomous agents. For the general public, it is like the closing press conference of a G7: you don't know in advance what will come out, but everyone knows the first sentence will count. Source : https://www.csis.org/programs/trump-xi-2026-summits
📊 Trend
This 24 September 2026 will go down as the day AI agents became a foreign policy subject in their own right. While Trump and Xi meet at the White House and treat AI as both a balance-of-power issue and a shared risk, a radical bill is filed in the US Congress to ban « super-AI » and create a dedicated department. In Europe, a thousand engineers meet at Station F to build the day-to-day tools. And meanwhile, money is pouring in: Basecamp receives 140 million to manufacture medicines with AI, Cyera collects 400 million to lock down the data that agents access. Alibaba prepares the next generation of its model and aims for the place of « operating system for Chinese devices ». For the general public, the message is clear: AI agents are no longer a trade-show subject, they have become a ground for negotiation between great powers, an investment market of several billion dollars, and a subject of law — all in the same day.