Sign Up | Collaborate with us Welcome back! Claude Fable 5 had been public for one day when Microsoft limited employee access while its lawyers read Anthropic's new data terms. The same Wednesday, Visa moved your card into ChatGPT, Mastercard launched a payment system for AI agents that spend fractions of a cent and Dario Amodei asked Washington for the power to block models like the one his company shipped Tuesday. In today's Generative AI Newsletter: Microsoft vs Fable: What did Microsoft's lawyers find in the new Claude's terms?
Visa x ChatGPT: What happens the first time ChatGPT buys the wrong thing with your card? Anthropic's proposal: Why would the most valuable private AI company ask to be regulated? Mastercard's machine money: What do AI agents buy with payments worth a fraction of a cent? Microsoft won't clear the new Claude for its own staff Claude Fable 5 went live Tuesday.
By Wednesday, Microsoft had moved to limit employee use of it while company lawyers decide whether to allow it at all. The dispute is the fine print we covered yesterday. Anthropic keeps every prompt and output that touches a Mythos-class model for 30 days, on every platform where the model runs, business customers included. Anything its classifiers mark as a possible policy violation stays for up to two years.
Anthropic's reasoning is that jailbreaks operate across many requests, so catching them takes a stored window of evidence. It says the data never trains models and gets deleted on schedule in almost all cases. Microsoft's lawyers read the same policy and saw customer data and confidential information sitting on another company's servers for a month. The review only covers the Mythos-class line, so Opus and Sonnet aren't part of it.
Fable is also included on paid Claude plans only through June 22, then moves to usage credits. That gives Anthropic eleven days to convince corporate legal teams the retention window is survivable, and the first verdict is Microsoft's.
You’re already three Claude releases behind. Most people haven’t noticed. Anthropic has been shipping new Claude capabilities almost every week. Skills, Connectors, Cowork, vibe coding.
Most professionals are still using Claude the way they used ChatGPT a year ago, missing the modes that actually change how work gets done. Outskill’s 2-Day Claude Mastery Workshop is a live walkthrough of every major Claude capability, taught by mentors who’ve spent 800+ hours testing them in real workflows. You’ll leave with working setups, not theory: Connectors automating your Notion and files, vibe-coded dashboards, and a clear sense of which Claude mode to use for what. Free this weekend.
Sign up free How the biggest names in AI get seen Updated time below Google, Nvidia, Oracle and IBM sit among the partners following this GTM strategy. The person who knows how companies at that level actually reach people is Steve Nouri, a Fortune 500 go-to-market advisor, and today he's handing over the playbook, live. Distribution is what separates the AI companies everyone knows from the ones nobody finds. Whether you build, back or run one, this is how the top players do it.
This is for you if: You want to put an AI product in front of millions, even if starting from zero. You run a company and need to win more of your customers' attention without paying for it. You back AI companies and want to spot the ones that can reach a market before you write the check. Ask Steve your questions live and leave with the same playbook that reached 14 million people.
It's the kind of access the big names usually keep in-house. It's live in a week and the room is filling up. Save your seat here. Your Visa card now works inside ChatGPT Visa plugged its payment network into ChatGPT on Wednesday.
Link your card and ChatGPT can search, compare and complete a purchase at any merchant that accepts Visa, which is most of them.
Forestell, Visa's chief product and strategy officer, gave the example on stage at Visa's payments forum in San Francisco. Ask for wireless headphones under $150 and it finds a pair and buys them. OpenAI already tried owning checkout alone. Instant Checkout launched late last year as a digital personal shopper, was error-prone enough to stall and charged merchants a fee most refused to pay.
OpenAI retired it in March. The new split hands each company the part it's good at. OpenAI builds the agents, and Visa runs authorization, fraud monitoring and disputes on the largest payment network outside China. Most purchases will start with the agent pinging you before it pays.
Forestell told AP how that ends. You approve a thousand times, and eventually the agent asks, "Do you want me to just not check?" ChatGPT has 900 million weekly users and only about 50 million pay, a gap we covered Monday when OpenAI started rebuilding ChatGPT into a superapp. This looks like one of the first steps needed. Neither company disclosed who earns what per purchase.
A cut of checkout from 900 million people is the kind of number an S-1 wants, and OpenAI's filing is already at the SEC. Anthropic wants AI regulated like airplanes A day after shipping the most capable model it has ever sold to the public, Anthropic asked the government for the power to ground models like it.
Amodei, Anthropic's CEO, made the case Wednesday in an essay and two policy frameworks the company says it will back financially. The model he wants copied is the FAA. Airplanes get tested and audited before they fly, and grounded when they fail. The framework reads like aviation law: Who it covers: models trained with more than 10²⁵ FLOPs of compute, built by companies with over $500 million in AI revenue or $1 billion in AI R&D spending. The test: mandatory third-party evaluation for four risks, biological weapons, cyberattacks, loss of control and AI that automates building better AI. The teeth: government power to block or reverse a release, with civil penalties tied to global annual revenue.
The states: no federal preemption of state AI laws unless Congress passes something at least as strong. His evidence is Anthropic's own product. Mythos Preview found thousands of high-severity vulnerabilities this year, including in every major operating system and browser, and Amodei writes that it proves AI models are now "tools of global and national strategic consequence." His position now is that transparency rules like the California law Anthropic helped pass are no longer sufficient. The why-now reads cleaner with the calendar open.
Anthropic filed to go public this month, and one federal rulebook is worth more to that listing than fifty state ones written in parallel. The thresholds also draw the line around companies Anthropic's size, so a mandatory government test is a cost it can absorb and a wall its smaller rivals get to climb later. Anthropic walks into its IPO as the company that already passes the test it designed. Lovable's agents can now buy what they build Tell Lovable to build you a flower shop and the agent writes the site, but a human still has to pull out a card for the domain, the hosting and the checkout.
Mastercard wants the agent to handle that part too. On Wednesday it launched Agent Pay for Machines, built for payments that are programmatic, always-on and sometimes worth fractions of a cent, and Lovable signed on as an early access partner alongside more than 30 names including Stripe, Coinbase, Cloudflare and Adyen. The mechanics are the product. Every agent carries a credential so the other side knows what's transacting, spending rules are enforced programmatically rather than on trust, and settlement is guaranteed across cards, accounts and stablecoins.
Mastercard's own example is the flower shop. One request and a defined budget, and the agent assembles the domain, images and payment pages as a chain of purchases across providers.
Lambert, Mastercard's chief product officer, calls the goal "a superbloom of AI business models," meaning services built for agents to buy from other agents at volumes and sizes no checkout page supports. Visa wired ChatGPT for shoppers and Mastercard built rails for the machines, on the same day. Both settle across cards and stablecoins, so the card networks are riding crypto rails while racing crypto-native players like Coinbase to own the trust layer on top. Their edge is the part a blockchain alone can't promise, which is dispute resolution, fraud coverage and your money actually arriving.
Nothing multiplies transactions like software that pays other software around the clock, and the networks earn a slice of each one. Tool of the Day: Mintlify Mintlify turns a GitHub repo into a documentation site built for both people and AI. You write pages in markdown, it handles the design, search and an assistant that answers visitors' questions from your content. It also supports llms.txt and MCP, so ChatGPT and Claude read your docs instead of guessing about your product. Anthropic runs the Claude API, MCP and Claude Code docs on it for over 2 million developers a month, and Lovable, from today's Mastercard story, runs its onboarding and support docs there too.
Try this yourself: Sign up free at mintlify.com and connect a GitHub repo. Start from the quickstart template and your docs site is live in minutes, written in plain markdown. Turn on the built-in assistant so visitors ask questions and get answers pulled straight from your pages. Run your existing docs through Agent Score at mintlify.com/score to see how readable they are for AI agents.
Who it's for: anyone whose product docs, API reference or internal knowledge base goes out of date faster than the team can fix it. Everything else you shouldn't miss Fable 5 flags basic biology as a risk: Two days after launch, users found the new model blocking plain questions about cancer and mitochondria, and Anthropic says it was "overly conservative" with the safeguards and is working to loosen them. A Redditor says he runs ChatGPT in his head: He claims he can predict the bot's answers without opening the app, and the thread split between mocking him and admitting they hear it too. China is squeezing a material the AI buildout needs: Beijing's export licenses for indium phosphide, the stuff inside the optical links between AI servers, keep stalling, and a 6-inch wafer now costs $5,000, up 250%. Stitch Fix's AI styling is working: The styling service grew active clients quarter over quarter on the back of its revamped AI shopping features, and says AI cut private-brand design from a multi-month cycle to about one week.
Learn more about AI from the experts building it 📸 Follow us on Instagram for fast, visual AI updates in 30 seconds. 🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business. 📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators. 🐦 Follow us on X for breaking AI news and real-time industry updates. 🧠 Learn how to build your next AI application with practical resources and expert guidance. 🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
Get in touch.