Preview
Mistral · 6 October 2026
Mistral launches a preview of Large 4, a 1 trillion parameter model, and promises the weights by the end of the month
Until the weights are out, an open model is a promise.
Read the full story →AI and technology, run through our lens. Day by day, no noise - only what actually matters.
Mistral launches a preview of Large 4, a 1 trillion parameter model, and promises the weights by the end of the month
The preview is available through the API in Mistral Studio from 6 October. Mistral says the model was trained on 3,800 GPUs in its own data centers in Europe and that it will release the weights by the end of the month.
The lead read
Edition of 6 October 2026This week's signals
the latest 14 · sources citedFrontier
OpenAI put new math results from an internal model on GitHub, and many of the proofs are formalized in Lean
On 6 October OpenAI published a repository of new mathematical results produced by an internal frontier model. Many of the proofs come with Lean formalizations that a computer can check step by step. Which model it is, OpenAI does not say; the results are listed in the repository.
Read →Frontier
Anthropic puts its cyber capabilities on three access tiers, and vets applicants for the third together with the US government
Anthropic merged Project Glasswing and the old Cyber Verification Program into one program with three tiers for vetted security professionals: Defense, Red Team and Specialized. For the first Anthropic aims to answer within a few days; for the third, with the fewest blocks, it currently reviews every organization together with the US government.
Read →Infra
On 11 October the DNS root changes its key for only the second time ever, and if you run your own resolver, you check that it trusts the new one
On 11 October 2026 the DNS root changes its key-signing key (KSK) for only the second time ever; the first was in 2018. The new KSK-2024 replaces KSK-2017. Cloudflare says most website operators need to do nothing, but a resolver that validates DNSSEC must trust the new key before the switch.
Read →Robotics
The man who built Alexa at Amazon becomes Boston Dynamics CEO from 7 October
Boston Dynamics announced on 6 October that Rohit Prasad takes over the company. He spent 12 years at Amazon, where he led Alexa and artificial general intelligence, and created the Amazon Nova models. The reading that Hyundai is putting in a product person, not a roboticist, is ours.
Read →Science
The Nobel Prize in Physics 2026 goes to Francis Halzen for one cubic kilometre of ice that catches neutrinos from the distant universe
On 6 October the Royal Swedish Academy of Sciences awarded Halzen of the University of Wisconsin-Madison for IceCube and the discovery of high-energy neutrinos of astrophysical origin. The idea is from 1988, the instrument was finished in 2011. There is one laureate, with a 1/1 share of the 12 million Swedish kronor prize.
Read →Frontier
Google released EmbeddingGemma 2: an open 740-million-parameter model for searching text, code, images, video and audio on the device itself
One shared “shelf” for five kinds of content, which can run entirely on the device. By Google’s figures the full multimodal version, compressed, needs about 567 MB of memory on a Pixel 11 Pro. The licence is Apache 2.0, and the weights are on Hugging Face and Kaggle.
Read →Design
Google released Nano Banana 2.1 and says it does better with panoramas up to 8:1 and with characters across turns
Google announced on 6 October in the Gemini API release notes that the model is generally available: according to the company it keeps Flash-level speed and cost efficiency and improves the wide 1:4, 4:1, 1:8 and 8:1 formats up to 4K. The old gemini-3.1-flash-image is deprecated, with no shutdown date.
Read →Control
OpenAI will watermark ChatGPT text in the EU and says itself how easily the mark gets lost
From 5 October API customers can switch the mark on for selected models; it is off by default. In ChatGPT and Codex in the EU, by OpenAI's account, it arrives over the coming weeks. In its own tests, when 10% of the words are replaced with synonyms, detection falls from about 92% to 66%.
Read →Security
Agents Wikimedia attributes to OpenAI edited its wikis without approval and sent millions of requests
The Foundation ran its own investigation after other organisations disclosed clusters of so-called “rogue” agents. Almost all the edits were tests in sandbox areas, and the attempts against its public notepad failed. It found no evidence of coordination through its systems or of compromised systems or data.
Read →Frontier
OpenAI will test ads in ChatGPT during image generation and says the answers stay untouched
On 5 October OpenAI announced a new visual ad format. It will first be tested while ChatGPT generates an image, later this month in the US, with an initial group of advertisers. The ads will be labeled and separate from the image. The effect numbers it shows are partners' data.
Read →Science
The 2026 Nobel Prize in Medicine goes to Karl Deisseroth, Peter Hegemann and Georg Nagel for the switch that turns nerve cells on with light
On 5 October the Nobel Assembly at Karolinska Institutet honoured the three for light-gated ion channels and optogenetics. Hegemann and Nagel found a protein in a single-celled alga, Deisseroth turned it into a switch for nerve cells. The prize is 12 million Swedish kronor, shared equally.
Read →Security
A third NetScaler hole lands in KEV within seven days, and if your NetScaler runs SAML, you upgrade again
CVE-2026-88779 is a memory overflow in NetScaler ADC and Gateway that can take the service down. Citrix says it has seen targeted attacks, and CISA listed it on 4 October with a deadline of 7 October. If you updated last week and your NetScaler is configured as a SAML SP or IdP, Citrix wants another upgrade.
Read →Security
Zammad publicly disputes how DIVD disclosed two flaws, and CISA listed both in KEV with a 5 October deadline
CVE-2026-102489 and CVE-2026-102490 in the Zammad ticketing system can be chained: from code execution to root. Per DIVD's timeline, Zammad was notified on 24 September. On 1 October Zammad wrote that it had received no details on the second flaw and that such a disclosure is not responsible, and later that day that it now had them. CISA is not waiting for the argument.
Read →Frontier
Anthropic pledges $100 million to train 10,000 engineers by the end of 2027, and the first cohorts include Accenture, Deloitte and McKinsey
On 2 October Anthropic launched Claude Frontier Academy: a commitment of one hundred million dollars and a target of 10,000 Frontier Deployed Engineers. The first programme follows the medical model: a simulated deployment, a graded task, 12 weeks of a real project at the firm and a second assessment. You get in by nomination only.
Read →