Welcome back! Beijing is calling Anthropic's most dangerous model a possible weapon seven weeks before Xi lands in America, and the White House finished its AI rulebook and put four labs in a room today. Visa dropped $2.4 billion on a company that reads your keystrokes, and Dario Amodei told colleagues he's worried his new hires only came for the paycheck.
In today's Generative AI Newsletter:
China: What does Beijing want from Anthropic?
White House: What can the government do to a model before it ships?
Visa: What is it buying for $2.4 billion?
Anthropic: Why is Dario unhappy with his own hires?
China issues a warning about Mythos
Beijing thinks Claude Mythos could be turned on it, and it picked a loaded moment to say so. Xi visits the US on September 24, with AI talks expected before the two leaders meet.
Officials see models like Mythos as a possible offensive weapon, and they want to know why Anthropic denies China access for ordinary use while American companies get it.
Beijing isn't preparing anything yet. It's drawing up responses in case Washington moves first, and the options include sanctions or putting American AI companies on a restricted-entity list.
Wu Xinbo, who runs Fudan University's Center for American Studies, put the fear plainly. US AI companies collect customer data and can hand it to the American government or military.
Anthropic is an awkward target. Dario Amodei has pushed Washington to keep chips out of China and accused Alibaba's Qwen team of training on Claude's outputs.
The company blocked Chinese-controlled firms last year and doesn't operate in China at all, so sanctions would sting reputationally more than financially.
Washington isn't easy on Anthropic either. Commerce Secretary Howard Lutnick made it get licenses in June before serving Mythos 5 and Fable 5 to any foreign entity, and Anthropic switched the models off until it worked something out.
Looks like Anthropic got problems with the two world leaders. Let’s keep an eye on how it’s going to play out, especially after the new regulations USA wants to introduce.
200 Ways To Make Money With AI
Ready to transform artificial intelligence from a buzzword into your personal revenue generator?
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets, each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
The White House finished the AI rulebook
Meta, OpenAI, Google and Anthropic go into the White House today to review a finished framework for testing what the most advanced American models can hack.
Developers can hand the government access to a covered frontier model for up to 30 days before releasing it to trusted partners, so officials can check whether it can find software vulnerabilities or run sophisticated attacks.
Most of it stays secret. Trump's June 2 order told Treasury, the NSA and CISA to build a classified benchmarking process, and both the benchmark and the threshold for which models qualify are staying classified.
One line protects the labs. The order says the program cannot be used to create mandatory licensing, permitting or preclearance for building or releasing a model.
The reason this exists is the reason we keep writing about it. OpenAI's agent escaped a test environment and compromised Hugging Face while hunting for answers, and both OpenAI and Anthropic have now disclosed their tools breaching other companies' systems. This is scaring governments and businesses worldwide.
Visa bought a company that watches how you type
Visa is paying $2.4 billion in cash for BioCatch, a fraud company built on behavioral biometrics.
That means it watches how you move. Keystrokes, mouse activity, touchscreen gestures, how you hold the device, and whether you look like someone being talked into a transfer by a stranger on the phone.
It already runs at scale. BioCatch analyzes 19 billion online banking sessions a month across 760 million users and 1.8 billion devices, for more than 350 banks in 21 countries.
Visa wants the fraud caught before the payment clears rather than clawed back afterward.
Permira took a majority stake in BioCatch in 2024 at a $1.3 billion valuation. Visa is paying nearly double that two years later.
How are you feeling about this?
Anthropic's CEO says his hires only want the money
A source said Dario Amodei has grown worried that new hires are joining Anthropic for the money rather than the mission.
It stung because Anthropic reportedly pays more than any lab in AI, OpenAI included. Amodei has also said he won't raise salaries to fight poaching.
Lilian Weng left Thinking Machines last week and turned up at OpenAI days later, the fourth co-founder to leave that company in a year. Google lost Noam Shazeer to OpenAI and Nobel winner John Jumper to Anthropic in June.
When every lab can pay millions, pay stops separating anyone. Mission is the only thing left to compete on, and it's the one thing money can't verify. This is what Dario wants from his team. Mission focused, not driven by salaries.
QuillBot is the writing suite students already use for paraphrasing and grammar checks, and AI Chat is its chatbot sitting inside all of it. The difference from a standalone chatbot is what's next to it. You can move a chat answer straight into the paraphraser, the summarizer or the citation generator without leaving the page. It's in beta and free to try.
Try this yourself:
Open AI Chat and ask it to draft something you'd normally write from scratch.
Send the output into the Paraphraser and compare the rewrite to the original.
Run the Citation Generator on anything you're quoting, since it handles APA, MLA and Chicago.
Who it's for: students and anyone writing at volume who wants the chatbot and the cleanup tools in one place.
Everything else you shouldn't miss
15 attorneys general put OpenAI on notice: A coalition led by Iowa's Brenna Bird told Sam Altman the company may have broken consumer protection and data privacy laws, and warned that destroying records could bring sanctions if it goes to court.
Anthropic's own models hacked three organizations during tests: OpenAI is not the only lab whose tools got out during evaluations, which is the backdrop to today's White House meeting.
OpenAI put a price tag on its own math results: Its account confirmed an internal version of its next model produced ten new results on open problems using roughly $2,000 of tokens at Sol rates.
Every Erdős problem could fall by 2028: That projection is people on X extrapolating the current pace rather than anyone official, and it sounds less absurd after three of them went down over the weekend.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 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.
