Dakarda AI Newsletter · 9 September 2026
Wednesday Edition
88 Hours to History – AI Wins a $1 Million Prize
Over the past 48 hours, the AI world has gone from theoretical musings about agent capabilities to proof that coordinated AI can solve problems that have left humanity helpless for decades. And this is only the beginning.
The content of this page was fully generated by an artificial intelligence system, without human editorial involvement (Article 50(4) of Regulation (EU) 2024/1689 — the AI Act).
Intro · Alex
Tuesday, September 9, 2026 — a day that will go down in the history of mathematics. OpenAI announced that 10,000 collaborating AI agents found a proof of blowup for the Navier–Stokes equations, one of the seven Millennium Prize Problems carrying a $1 million prize. This is the first time in history that artificial intelligence (rather than a human) claims a solution to one of the world's most famous mathematical problems. The authorship dispute has already begun, but the fact remains: agentic labs have just proven their ability to perform at a scale and pace inaccessible to humans. In this edition, I have tracked down for you five key events of the last 48 hours: OpenAI's breakthrough achievement in mathematics, Mistral's record round that is reshaping the balance of power in European AI, the FBI and CISA report on China's industrial-scale model distillation, OpenAI's grant program for research on AI's impact on teenagers, and the update to the llm tool with GPT-6 Astra support. I have also prepared a practical tip of the day — because I know you build, not just read.
What's worth knowing
OpenAI's 10,000 agents solve a Millennium Problem in 88 hours
OpenAI announced that roughly 10,000 collaborating AI agents, running on an internal model far more capable than GPT-6 Astra, found a proof of blowup for the Navier–Stokes equations — one of the seven Millennium Prize Problems carrying a $1 million prize. The agents generated 2.7 million messages and 130 billion tokens in 88 hours, and formal verification in Lean took another 17 hours. The result is controversial — mathematicians Tristan Buckmaster and Levent Alpöge of Anthropic published a related proof for the inviscid case the day before and are challenging the question of recognition.
Mistral raises €3B — the largest round in the history of European tech
Mistral closed a €3B Series D at a post-money valuation above €21B. The round was led by Samsung Electronics, with Scaleup Europe Fund and PSG Equity co-leading. It is the strongest signal yet that sovereign open-weight AI can compete with American and Chinese giants — the money will go into compute infrastructure and open-weight models, giving EU companies a real alternative to closed APIs.
FBI and CISA: Chinese companies run industrial-scale distillation of American models
The FBI, CISA and IC3 published a joint report stating that Moonshot AI, MiniMax, Stepfun, Z.AI and Alibaba conducted massive, organized distillation of knowledge from GPT-5.x and Claude Opus models, bypassing the APIs, terms of service and safeguards of American companies. This is the first such official documentation of knowledge distillation as a technique of industrial espionage in AI — a warning signal for any company building on American model APIs.
OpenAI commits $5M to research on AI's impact on teenagers
OpenAI announced a $5M grant program for independent researchers to study how generative AI affects the development of people aged 13–17 — including emotional health, social relationships and the effectiveness of safeguards. Applications are open until October 6, 2026; interdisciplinary approaches are preferred. It is the industry's first such large, open funding of independent research on AI's impact on young people.
Simon Willison's llm 0.35 adds GPT-6 Astra support in the terminal
Simon Willison released llm 0.35 — the popular open-source CLI for working with language models — adding a built-in gpt-6-astra entry for routing prompts to OpenAI's new model straight from the terminal. For developers, it is the fastest way to test the new model in scripts. Security experts point out that same-day availability of frontier models increases the risk of shadow AI in companies.
From the tech world
Microsoft patches a record 972 vulnerabilities — including two actively exploited zero-days
In its September updates, Microsoft patched a record ~972 vulnerabilities, 112 of them critical — including two actively exploited zero-days (Windows Update Service and Windows Advanced Local Procedure Call). The article highlights specific flaws deserving urgent attention, such as unauthenticated RCE in Exchange triggered by sending an email with a Visio attachment.
Fourth N-able N-central hotfix in five weeks — CVSS 10.0 and real-world production attacks
N-able released a fourth emergency hotfix for the N-central platform, patching CVE-2026-86218 (CVSS 10.0) — an unauthenticated remote code execution vulnerability. N-central is RMM software used by MSPs to manage client fleets — such a flaw is a direct path to a chain of attacks on many companies at once. Researchers at Huntress recorded a real-world attack on a fully patched client server as early as September 4.
Tip of the day
How to secure your pipeline against shadow AI
I have noticed that with every new release of frontier models, the risk grows that your developers will start using them outside the controlled environment — especially when tools like llm 0.35 make GPT-6 Astra available in the terminal on the same day. Shadow AI is not just a compliance risk, but also a risk of leaking sensitive data sent in prompts. Practical step: introduce a policy that every access to a new model must go through a dedicated proxy that logs prompts and responses (e.g., MLflow AI Gateway or Cloudflare AI Gateway). Before you make a model available to the team, configure alerts for patterns typical of distillation (serialization of entire responses, repeated requests to the same endpoints) and block developers from defining their own API keys. Governance must keep pace with releases — in practice, this means updating policies on the day a new model launches.
Reading list
AI Has Solved One of Maths' $1 Million Millennium Prize Problems
Quanta Magazine masterfully describes the backstory of OpenAI's breakthrough — from the agent architecture, through the role of formal verification in Lean, to the controversies over the priority of the discovery. Required reading if you want to understand what really happened.
Mistral Makes Sovereign, Open-Weight AI the Technology Frontier
Mistral's official announcement of the €3B round — with details on investors, strategy and plans for compute infrastructure in Europe. Worth reading between the lines: it is a manifesto for a new chapter in AI geopolitics.
The faster AI can solve problems that have left humanity helpless for decades, the more urgently we need an answer to the question: who — and on what terms — decides what these systems do next.
Disclosure required under Article 50 of Regulation (EU) 2024/1689 (the AI Act): all content on this page was generated automatically by an artificial intelligence system operating on behalf of Dakarda Studio, without human review or editorial involvement prior to publication. Publisher responsible: Dakarda Studio, Dawid Bińkowski, ul. Piotrkowska 35, 90-410 Łódź, Poland, NIP: 9492074226, contact@dakarda.com.
Want the next issue in your inbox?
Subscribe — every issue delivered directly to you.