Mistral CEO Arthur Mensch and Anthropic CEO Dario Amodei. Nathan Laine/Bloomberg via Getty Images; Jason Henry/Bloomberg via Getty Images The US imposed export controls on Anthropic's newest AI models. In response, Anthropic closed access. The restrictions could play directly into Mistral's long-standing AI sovereignty pitch. Mistral's CEO has warned for more than a year that Europe risks becoming dependent on US AI firms. The White House's move to restrict access to Anthropic's newest AI models could hand one of Europe's leading AI startups exactly the opening it has been preparing for. On Friday, US officials imposed export controls on Anthropic's cybersecurity-focused models, Mythos 5 and Fable 5. They cited national security concerns that safeguards meant to prevent misuse of Fable 5 could be bypassed. The restrictions block any foreign national from accessing the two models. Anthropic responded by suspending access to the systems altogether, creating uncertainty around who ultimately controls access to frontier AI. The episode has the hallmarks of a scenario Mistral's CEO, Arthur Mensch, has been warning about for more than a year — and has since made part of his pitch for why people should choose it over models from US firms like OpenAI and Anthropic. Get up to speed on Anthropic's latest drama with the White House with this cheat sheet. Speaking at London Tech Week in June 2025, Mensch warned about American AI companies "having the keys" for their models, adding that he sees it as European companies "giving leverage to their providers." "At some point, you need to be able to turn it off or turn it on, and you don't want to leave it to another country," he said. Sovereign AI Since then, the French startup has positioned itself around AI sovereignty, the idea that governments and businesses should not become overly dependent on a handful of foreign AI providers. To do so, it has championed open-weight models that customers can deploy on their own infrastructure and customize using their own data. "European governments are coming to us because they want to build the technology and they want to serve their citizens," Mensch said on the "Big Technology Podcast" in January. Mensch doubled down on his position at a hearing on digital sovereignty and AI at France's National Assembly last month, warning that Europe has two years to build its own artificial intelligence infrastructure before becoming permanently dependent on American tech giants. At Mistral's first-ever summit in Paris that same month, executives, government officials, and enterprise customers repeatedly returned to the same theme: AI sovereignty. "You need to know where your data is and what happens to your data," Jan van den Bremen, Accenture's technology lead across Europe, the Middle East, and Africa, told Business Insider at the event. While Mistral is widely considered the top AI model provider in Europe, it lags behind the likes of Anthropic in terms of valuation, model capabilities, and user numbers. The French startup was last valued at around $13.6 billion, compared to Anthropic's $965 billion valuation. But the Anthropic restrictions make Mistral's core argument easier to understand: Most of the time, control of AI ultimately lies with the provider and its government. Read the original article on Business Insider

Lachlan Murdoch and Rupert Murdoch's Fox is buying Roku. But why? Adrian Edwards/GC Images There are so many things to stream. Wouldn't it be great if you could see them all in one place? And if you owned the place everyone came to when they wanted to stream, wouldn't that be worth a lot of money? That's the logic behind the Fox/Roku deal. Except … Rupert Murdoch chased the TV Guide of the Future for years. He spent billions trying to build an on-screen map for digital television before giving up. Now his son Lachlan looks like he's making it happen by buying Roku in a $22 billion deal. On paper, Lachlan's transaction seems like the thing his dad spent much of the '90s and 2000s trying to put together: Once the deal closes, the Murdochs will own the interface that 100 million households use when they want to stream something. It's a big deal because it means Fox won't just be a content company, but a distributor as well. Buying Roku gives Fox real estate it can use to promote its own services, like Tubi, its free streamer, and Fox One, its paid streamer. But it can also sell subscriptions to competitors' services and get paid when it does. And it can sell ads on all of the above. One big catch: Fox is spending $22 billion to own a TV Guide. But it's not the only TV Guide. Unlike the '90s and 2000s, there are now lots of ways to get something you want to watch onto a screen of your choice. Roku might be your home screen. But you might also use YouTube, or Apple, or Amazon, or Samsung, or Walmart's Vizio. You might still have a cable TV subscription, which would mean your cable TV provider is your home screen. Maybe you never use any of those services, and just turn on Netflix and never spend time anywhere else. The other big catch: In the internet age, owning a portal that millions of people use to access stuff they like gives you lots of power. But it doesn't give you all the power: The biggest content companies and app makers have their own leverage. Apple can't sell an iPhone that doesn't let you use Instagram, and Roku can't be a TV portal if it doesn't include YouTube and Netflix, the two biggest names in streaming. You could see how that gave YouTube and Netflix leverage when Roku went public in 2017: Back then, the company disclosed that Netflix generated about a third of its viewing but immaterial revenue, while YouTube, its most-watched ad-supported app, generated no revenue at all. The details have changed a bit since then — I've asked Roku for the latest — but the lesson hasn't: If you are big enough, you don't pay the same tolls as everyone else. Let's be clear: Lachlan Murdoch is getting value for his $22 billion. Roku is a massive player in streaming, and it makes $4.7 billion a year by selling ads and subscriptions. Owning it gives Fox a way to diversify its revenue streams: Now Fox can make money from its own shows and services, but it can also get a piece of the business happening around other people's shows and services. That's a position lots of media companies have wanted to be in for a long time. It doesn't always work out — ask HBO, which keeps getting combined with distribution companies (like Time Warner Cable, AOL, AT&T) and then sold to someone else when that doesn't work. It also has a built-in tension. Yes, owning a distribution platform means you can give your own stuff a boost — that's why Roku has three different streamers of its own. But if you turn on your Roku TV and want to stream "The White Lotus," you don't want to wade through a bunch of ads for Howdy or the Roku Channel before you get there. So Roku, like every other distributor, has to balance out the stuff it wants you to see vs. the stuff you want to see. That kind of tension, by the way, might normally be an issue for regulators to examine. But in 2026, the odds that a Fox-friendly White House is going to stop the Murdochs from getting a deal they want are very low. Just ask Larry and David Ellison, whose Paramount got the go-ahead to buy Warner Bros. Discovery from Trump's Department of Justice on Friday afternoon — and then hosted a UFC fight night on the White House lawn two days later. All of which makes the Roku deal a big event — but not one that completely remakes streaming. When Rupert Murdoch wanted to own the next-generation TV Guide, he imagined owning the only way to get to TV. But there are lots of TV Guides in 2026. Read the original article on Business Insider

The only plausible response to videos of aliens on television, at this point, would be cries of “that’s AI,” “fake,” and propaganda flowing in all directions.

Foundation models are reshaping computational biology. Pretrained on massive corpora of protein or genomic sequences, models such as ESM2 (a protein language... Foundation models are reshaping computational biology. Pretrained on massive corpora of protein or genomic sequences, models such as ESM2 (a protein language model) and Evo 2 (a DNA language model) capture statistical regularities of biological sequences. These transfer well to a wide range of downstream tasks, including structure prediction, variant effect, and functional annotation. Source

Jonathan Stempel / Reuters: A US judge dismisses xAI's lawsuit alleging OpenAI stole trade secrets, saying xAI failed to show OpenAI induced a former xAI engineer to divulge trade secrets  —  A federal judge on Monday dismissed a lawsuit by Elon Musk's artificial intelligence company xAI that accused rival Sam Altman's OpenAI …

AMD's stripping of TSME from consumer CPUs appears to be a deliberate, covert move.

The UK’s proposed under-16 social media ban could bring stricter age checks, teen defaults, and feature limits for social and AI platforms. The post UK Weighs Under-16 Social Media Ban for 2027 appeared first on TechRepublic.

Ben Thompson / Stratechery: Anthropic's belief in its own commitment to safety gives Anthropic the license to aggressively favor its business and even challenge the US government  —  I'm sympathetic to the cynics who consistently characterize Anthropic's public statements, particularly those surrounding their model releases …

Discussion | Link

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-agent/releases/tag/0.3a0">datasette-agent 0.3a0</a></p> <blockquote> <ul> <li>New tool, <code>execute_write_sql</code>, which requests user approval and then writes to a database - taking user permissions into account. <a href="https://github.com/datasette/datasette-agent/issues/27">#27</a></li> </ul> </blockquote> <p>I added a mechanism for asking user approval in <a href="https://simonwillison.net/2026/Jun/10/datasette-agent/">datasette agent 0.2a0</a>. The new <code>execute_write_sql</code> tool can now prompt the user for all kinds of useful operations. Here's an example where I add some pelican sightings to my <code>pelican_sightings</code> table:</p> <p><img alt="Screenshot of a chat interface showing a write SQL confirmation dialog. User message (blue bubble): &quot;I saw 4 pelicans flying over the harbor&quot;. Collapsed tool section: &quot;► Tool: execute_write_sql&quot;. A yellow-bordered confirmation card reads: &quot;Confirm write SQL batch / Database: pelicans / Statements execute in order. If one statement fails, later statements will not be executed. / Statement 1 / INSERT INTO pelican_sightings (number_of_pelicans, notes) VALUES (:number_of_pelicans, :notes); / number_of_pelicans 4 / notes Flying over the harbor&quot;. A table with columns &quot;Operation, Database, Table, Required permissions&quot; shows row: &quot;insert, pelicans, pelican_sightings&quot; with permission buttons &quot;insert-row&quot;, &quot;update-row&quot;, &quot;delete-row&quot;. Below: &quot;Execute 1 write SQL statement against database 'pelicans'? / Asked by tool: execute_write_sql&quot; with &quot;Yes&quot; (blue) and &quot;No&quot; (gray) buttons." src="https://static.simonwillison.net/static/2026/i-saw-pelicans-agent.jpg" /></p> <p>The new version also enhances the <code>datasette agent chat</code> terminal mode to support approvals, and adds several new options including <code>--unsafe</code> mode for auto-approving them:</p> <blockquote> <ul> <li><code>datasette agent chat</code> can execute tools that require user approval. <a href="https://github.com/datasette/datasette-agent/issues/30">#30</a></li> <li>Three new options for <code>datasette agent chat</code> - <code>--root</code> to run as root, <code>--yes</code> to approve all ask user questions, and <code>--unsafe</code> for both.</li> <li>Tools can now provide plain text alternatives to HTML, for display in the <code>datasette agent chat</code> CLI. <a href="https://github.com/datasette/datasette-agent/issues/31">#31</a></li> </ul> </blockquote> <p>The <code>datasette agent chat content.db -m gpt-5.5 --unsafe</code> command can now be used to chat directly with a specific database <em>and</em> directly modify it through prompts like "create a notes table", "add a note about X" etc.</p> <p>Tags: <a href="https://simonwillison.net/tags/projects">projects</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/annotated-release-notes">annotated-release-notes</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/llm-tool-use">llm-tool-use</a>, <a href="https://simonwillison.net/tags/datasette-agent">datasette-agent</a></p>

Natalie Breymeyer / Axios: Radical Numerics, which is developing AI models that learn directly from biological data, raised a $50M seed led by Emergence Capital  —  Radical Numerics, an AI research lab for biological data, raised a $50 million seed round, CEO Eric Nguyen tells Axios.

The enterprise AI pitch cycle follows a familiar arc. A vendor or internal team wheels in a demo. The agent performs: it retrieves records from three separate systems, synthesizes a […] The post Why Most Agentic AI Projects Fail in Production appeared first on BigDATAwire.

Google discovered and disrupted the sprawling campaign, which stole RedCAP credentials to breach numerous institutions and exfiltrate sensitive data.

Executive leaders may not be saying it aloud, but business objectives and priorities don't always promote timely disclosures.

Mixture-of-experts (MoE) models have quickly become a foundational component of modern, large-scale AI systems. They are widely adopted because they enable... Mixture-of-experts (MoE) models have quickly become a foundational component of modern, large-scale AI systems. They are widely adopted because they enable substantially larger model capacity while activating only a subset of parameters for each token, offering an unparalleled approach for scaling performance within a practical compute budget. As model scales continue to grow… Source

Discussion | Link

Remembering the ups and downs of the Intel Mac era as it finally winds down.

Increase productivity with your LLMs The post How to Effectively Align with Claude Code appeared first on Towards Data Science.

Will the Sun roast Earth’s plants or starve them?

Discussion | Link