This page is available in English: Home, Bio, Work, Blog, Contact, Privacy, Photo rights and the ticker. Some images may still contain Dutch text.
Archive From the archive of The Latest A.I. News
This is an archive page with the A.I. news of Wednesday 30 September 2026. This day was compiled afterwards from reports that appeared at the time, so it was not kept on the day itself; a lot may have changed since. Today's news is on The Latest A.I. News.
Compiled afterwards on
Google releases Gemini 4 Argon to a small group of security partners first, and the White House unveils a voluntary AI pledge. Elsewhere, new agents arrive at DoorDash and Instagram and ElevenLabs lets staff sell shares.
The new model is aimed at coding, research and writing but stands out for its cybersecurity skills. For now Google gives access only to partners in its Fairwind program; the company says it can find, verify and fix serious software flaws with little human direction. Google claims higher scores than GPT-6 Astra and Anthropic's models on several benchmarks.
The document, the Joint Commitment on Frontier Responsibilities, asks developers to make sure their systems work as intended and to fix problems quickly, for example with an independent oversight board. It is voluntary, so breaking it carries no penalty. The photo Trump shared shows Unites States misspelled; a separate executive order tells agencies to say superintelligence or SI instead of AI.
Sam Altman presented the API at DevDay as a limited preview. Developers supply a list of options and the Luna model returns a probability for each, which makes it very fast. It resembles Jev from TypeSafe AI; one demo used such a model to check agent actions at $2.94 per run versus $372 with a large language model.
Employees can cash in part of their vested shares in a $300 million sale, with large investors such as Wellington and T. Rowe Price buying. The valuation is double that of the February funding round. It is the second such sale, and startups use these to keep staff from leaving.
A journalist reported that the Muse agent could read his Messages app on his Mac even though he thought he had switched off full disk access. Meta says the integration is opt-in and both that access and the Messages connector must be enabled. A Meta executive does concede that Muse gave an incorrect explanation; distrust lingers because of Meta's earlier privacy problems.
Users can simply text requests such as their usual order and the agent works out which past order is meant. It proposes a cart, can send photos of dishes and accounts for dietary needs in group orders. There is a waitlist in the US.
The chatbot looks at figures such as views, retention, likes and comments and looks for patterns in how videos perform. It focuses on analysis rather than making content. Usage is capped, with higher limits for Meta One subscribers.
A toggle lets users search by text or voice, after which matching filters appear and wish-list properties can be compared with AI summaries. The update also brings maps of contacts' trips and new services such as meal delivery. The CEO deliberately avoids a chatbot-style interface.
According to an opinion piece, only about 2.2 percent of consumers paid for AI in May, at roughly $31 a month on average. Even with as many subscribers as Netflix, revenue would stay below a third of OpenAI's operating costs. That is why companies are shifting to business customers, although Meta's ad income lets it wait longer.
Every day has its own page. They stay up for three months.
Everything here is a summary of reports from other media, not original research. The label (Certain, High, Medium, Low) is my estimate of how reliable the source is. The summaries are generated automatically with AI and may contain mistakes; read the article itself for the details (the star marks the leading source). Sources: leading A.I. and tech sites and the official blogs of the A.I. companies.
