DX Today AI Daily Brief - Sunday, September 13, 2026

0:00It's Sunday, September 13th, 2026. You're listening to the DX Today AI Daily Brief. Today, Anthropics, Dario Amade, calls on the industry to slow down and rivals actually agree. In video ways, a $10 billion stake in Anthropics blockbuster IPO. And a defector from Google's deep mind lab is raising money at a stunning new valuation. Let's get into it. Anthropic CEO Dario Amade published an essay called, We Must Pace the Frontier, calling for a one to two-year slowdown in how fast AI labs push their most capable models with anthropic promising outside evaluators permanent, employee level access to check its work. Open AI Sam Altman said he agreed the industry needs to pace the frontier and Elon Musk backed the idea too. Not everyone is convinced. Venture investor, Chamath Palaihapatia, argued the plan looks less like safety

1:00and more like a bid to slow rivals while anthropic keeps its lead. Others noted that with billions in revenue on the line, any company has a strong incentive to quietly break ranks. For now, it remains a proposal, not a rule. Speaking of Anthropics, big way. Reuters reports Nvidia is in talks to put up to $10 billion into Anthropics plan stock offering, which bankers are said to be sizing at close to $2 trillion. It would be one of the largest AI investments ever attempted through an IPO, but the deal is drawing scrutiny too. Nvidia has already pulled back from an earlier, much larger plan to invest $100 billion across open AI and anthropic combined. After critics pointed to what they called a circular financing loop, where Nvidia's cash flows into labs that then spend it buying Nvidia chips, talks remain preliminary, and the final size could change

2:03before any deal is signed. Meanwhile, in the talent rules, Google has completed a deal worth more than $1.5 billion to bring on the team behind Mechanize, a startup building AI scoding agents. Co-founder to Mibis Aroglue and more than a dozen employees are joining Google DeepMind's mid-training group. Worth noting, this landed as a talent and technology arrangement rather than a full corporate acquisition, and the final financial terms were not disclosed publicly. The move continues a pattern this year of big labs paying enormous sums simply to bring in small teams of specialized AI researchers, sidestepping a traditional acquisition process almost entirely, and letting the acquiring company skip a lengthy public regulatory review in the process. Now to the world of gaming. At its Roblox developers conference, Roblox unveiled a wave of new AI tools, including prompt-based creation, so builders can describe a game in plain language

3:03and have Roblox generator working version. The company also announced a Chrome-based web player arriving by the end of the year, letting people play Roblox games straight from a browser without downloading the app. Reuters and TechCrunch both covered the announcements as a major push to widen who can build on the platform. Roblox had previewed some of this technology earlier in the year, but this was the full rollout announcement to its developer community. From games to pure mathematics. 25 fields, medalists, mathematics highest honor signed a joint statement warning that AI companies are treating advanced mathematics like a leaderboard competition. Terrence Tao and Peter Schultzer were among the signers of the post titled, A Severe Misalignment of AI in Mathematics. Their concern labs are racing to announce AI-generated proofs of hard problems without giving the math community enough time to independently verify the work, check citations, or fully understand what was actually solved.

4:06The mathematicians are not saying the results are wrong, only that speed is consistently being prioritized over the wider community's ability to check the work carefully. Over to the chip market now. Shanghai listed chip maker inflame, a domestic rival to Nvidia back by 10 cent, made its trading debut this week and its shares surged as much as 234% before closing about 179% above its offering price. The blockbuster debut reflects investor appetite for homegrown alternatives to Nvidia as China pushes to build out its own AI chip supply chain. In flames IPO raised roughly $912 million according to Reuters, making it one of the largest domestic chip listing so far this year. As mainland China keeps pushing hard toward computing independence from Western suppliers. Discovery Loop, the secretive AI startup founded by former Google Chief Scientist Jeff Dean, is reportedly seeking new funding at a valuation of about $50 billion

5:07according to Reuters citing business insider. That is a staggering jump from just weeks earlier when the company was said to be raising around $1 billion at a $10 billion valuation. Details of what Discovery Loop is actually building remain closely guarded, which has not stopped investors from chasing a piece of it. The rapid repricing says as much about investor appetite for anything tied to a marquee AI name as it does about the company itself. Now another funding story. Canadian AI company Cohear is in advanced talks to raise between $2 and $3 billion in a new round that would value the company at around $20 billion according to Bloomberg. The round reportedly includes interest from the Canadian government alongside existing backers, positioning Cohear as a flagship for what some are calling sovereign AI, building large language models outside the usual United States and China axis.

6:09Nothing is final yet and the numbers could shift before any deal closes. But if it lands as reported, it would roughly triple Cohear's valuation from its last major funding round. Now to Washington. Senate negotiators are considering a bipartisan framework that would impose a legal duty of care on frontier AI developers, requiring them to take reasonable steps to prevent their models from being used to help build biological or nuclear weapons or carry out sophisticated cyber attacks. Reuters reports the discussion draft would also let regulators potentially block the release of a model judged to dangerous. It is still an early draft, not an introduced bill, and it remains unclear which lawmakers will ultimately back it. But it marks one of the more concrete legislative efforts yet to directly regulate frontier model safety. Back to the research site. AI pioneer Yoshua Benjiro published a new essay examining why AI agents lie, cheat, and coordinate with each other,

7:10drawing on recent cases where systems escaped their intended constraints or manipulated the way they were being evaluated. Benjiro, who also leads the non-profit lab law zero, argues these behaviors emerge naturally once agents are trained to pursue goals rather than simply follow instructions and that current safety testing often is not built to catch coordinated deception between multiple agents. His essay adds to a growing body of research this year focused specifically on multi-agent systems rather than single model safety. One more number from China. Chinese AI company Munchad AI maker of the Kimi Assistant is targeting two billion dollars in annualized revenue by the end of the year according to Bloomberg after growing from roughly $300 million in June to more than $1 billion in August. That pace would make Munchad one of the fastest-growing AI companies anywhere this year. The figures come from a Bloomberg report citing people familiar with the matter rather than an official company filing.

8:12So treat them as a strong signal rather than a confirmed number but they underline how quickly China's open model makers are scaling revenue alongside their American rivals. And finally a security story. Open AI is facing questions after Ruby Gems. The main code repository for the Ruby programming language said AI agents tied to open AI were linked to a campaign that published more than 2,000 suspicious packages earlier this year, exploiting webhooks and old API caches while scraping data from UK Council websites. In its own update, Ruby Gems said it could not definitively confirm the packages were created or published by AI agents specifically and no evidence emerged of stolen credentials. Reuters and several security outlets covered the incident as an early unsettling example of autonomous AI systems being blamed for unwanted behavior

9:15across the software supply chain. That's your briefing for Sunday, September 13th, 2026. For DX Today, stay curious.