Daily Briefing
19 September 2026 · 5 stories
🔥 Top story
1Anthropic readies a brand-new AI model just before its stock market debut
Anthropic, the U.S. lab behind Claude, is preparing a brand-new AI model that should be released before its stock market listing this autumn. The stakes are simple: regain the lead over OpenAI, whose latest model GPT-6 Astra overtook Anthropic for the first time in two and a half years on the OpenRouter platform. Concretely, businesses now spend 13% of their AI budget on Astra versus 8% on Claude. Anthropic is already valued at 965 billion dollars and earns more per year than its rival (65 billion in annualised revenue versus 40). For the general public, it is a clear signal that the race between the American AI giants is still wide open, and that Anthropic's upcoming IPO will be one of the financial events of the year. Source : https://www.aol.com/articles/exclusive-anthropic-considers-releasing-ai-000506000.html
2Anthropic hands Accenture a billion-dollar safety check-up
Anthropic has signed a one-billion-dollar five-year deal with the consulting firm Accenture to have its most powerful AI models tested by outside experts, away from the lab. Picture a car maker asking an independent body to crash-test its vehicles before they hit the showrooms. The goal is to spot the flaws, the biases and the unexpected behaviours, then publish the results. It is the first time a lab of this size has outsourced the auditing of its top models in such a systematic way. For businesses that use Claude in sensitive areas (healthcare, banking, defence), this kind of outside check-up becomes a strong selling point. Source : https://www.techinasia.com/news/anthropic-accenture-1b-ai-safety-deal
3Google, NVIDIA and 18 electricity giants team up to free up data-centre power
A coalition led by Google, NVIDIA, Anthropic and 18 large electricity operators (including AES, Constellation and National Grid) has launched the AI Energy Management Alliance. The aim is to make data centres far more flexible on the power grid. Concretely, the idea is to be able to slow a data centre down for a moment when the grid is under strain — much like turning off the air-conditioning in an office block during a heatwave — and avoid building new power plants. Estimates point to roughly 100 gigawatts of capacity freed up, the equivalent of several dozen nuclear reactors. For ordinary households, this means the AI boom should not, in theory, lead to power cuts or huge electricity bill hikes. Source : https://www.businesswire.com/news/home/20260916005000/en/Emerald-AI-Launches-AEMA
4A London start-up beats humans at forecasting thanks to AI
Mantic, a London start-up founded by a former Google DeepMind researcher, has raised 25 million dollars in a round led by Radical Ventures, with Microsoft's fund (M12) and Mira Murati's lab (Thinking Machines) joining in. Its twist: its AI models do not answer questions, they make numbered predictions. At the Metaculus Cup 2026, Mantic beat the 676 human forecasters, including on the Colombian presidential election, called correctly one week before anyone else. For the general public, this is the arrival of a new kind of agent: no longer one that writes or chats, but one that predicts the future better than a human specialist. Hedge funds are already on the call. Source : https://mantic.ai/blog/series-seed-announcement
5A new kind of AI, faster and 444 times cheaper, can play Doom
TypeSafe AI, an American start-up founded by a former OpenAI researcher, has raised 40 million dollars and unveiled Jev, a radically different AI model: instead of generating text like ChatGPT, it returns a structured decision — yes or no, a category, a score. The company claims its system is 40 to 200 times faster and up to 444 times cheaper than the big conversational models (numbers claimed by the company). Within 36 hours, 140,000 people had joined the waiting list. The most striking demo: Jev plays Doom — the 1990s shooter — in real time, making decisions frame by frame. For the general public, this is the announcement that a new category of AI, specialised and far more efficient, is emerging alongside chatbots. Source : https://techcrunch.com/2026/09/18/typesafe-jev-system-one-model/
📡 To watch
Trump and Xi to discuss AI 'guardrails' at a dinner on September 24
On board Air Force One on Friday, Donald Trump confirmed he would discuss international cooperation on AI 'guardrails' with Xi Jinping, without specifying which ones. The Treasury Secretary mentioned 'good practices against non-state actors'. The state dinner on September 24 will also bring together the heads of OpenAI, NVIDIA and Qualcomm. Watch closely: the joint final statement. Source : https://thehill.com/homenews/administration/5467897-trump-xi-ai-guardrails
Anthropic automates 26% of its own research with Claude
Anthropic has launched a life-sciences programme and a controlled wet-lab to test Claude in real-world conditions. The striking number: Claude already drives 26% of the lab's own R&D. The company is offering up to 50,000 dollars in credits to academic researchers who use its tools. The first scientific results are expected before the end of 2026. Source : https://www.anthropic.com/news/life-sciences-program
Teladoc consolidates medical chatbot agents on its platform
Several signals point to a consolidation of medical chatbot agents on the Teladoc telehealth platform, notably through Doctor On Demand. To be confirmed in the next two weeks, after the Trump-Xi summit. No consolidated official statement yet. Source : https://www.teladoc.com/news/
Anthropic's IPO draws closer
The S-1 form was filed confidentially on June 1, 2026, valuing Anthropic at 965 billion dollars. The IPO window opens in mid-October or after the November mid-term elections. For the general public, this would be one of the biggest tech listings ever. Source : https://apnews.com/article/anthropic-ai-claude-ipo-572bb6cc12053c7aa95f775285cf4b73
📊 Trend
This Saturday shows a double movement. On one side, U.S. labs (Anthropic, OpenAI) keep accelerating, with new top-tier models and historic fund-raisings. On the other side, regulators and big business are taking the lead: external audits worth a billion dollars, coalitions to make data centres compatible with the power grid, diplomatic summits between Europe, the United Kingdom, the United States and China. For the general public, this confirms that AI is no longer just a technical subject: it has become a topic for chanceries, trading floors and electricity bills. The convergence between Wall Street, the White House and Brussels on the need for 'guardrails' is probably the headline of the week.