Collaborate with us Welcome back! Claude found weaknesses in encryption that human experts missed for years, and 1,132 of the people building frontier AI asked Washington to slow their own employers down. Meanwhile Trump closed the border to Chinese robots and Microsoft unveiled an AI that plays attacker and defender at the same time. In today's Generative AI Newsletter: Anthropic: Is Mythos really that good?
The letter: Why did OpenAI and Anthropic employees ask the USA to slow their companies down?
Mythos poked holes in unbreakable encryption Encryption is the math keeping your bank login and your messages private, and it holds for one reason. The world's best mathematicians attack it for years and come up empty. Anthropic pointed Claude Mythos at two of its most studied pieces, and Mythos found gaps in both. HAWK is a signature scheme built to survive quantum computers.
Human experts reviewed it for two years and passed it twice, and Mythos cut its strength in half in 60 hours. It also made the best-known attack on a weakened version of AES, the cipher behind most internet traffic, 200 to 800 times faster. Claude turned the AES job down at first, insisting there was nothing left to find. Three days of pep talks and a billion tokens later, it had the attack.
Plenty of ciphers guarding real systems have had far less review than these two, and Anthropic's own researchers needed a month to verify a week of Claude's work. Nothing you use today breaks. Both were research targets. 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. Get The Free Guide 1,132 AI insiders asked Washington to hit the brakes The letter is called Pacing the Frontier and it went public yesterday with 1,132 signatures from inside OpenAI, Anthropic, Google and Meta. It asks the US government to support an international effort to deliberately pace frontier AI development.
In plain words, build the machinery to slow every lab down at once, because no lab can afford to slow down alone. Anthropic co-founders Jared Kaplan, Jack Clark and Ben Mann signed, along with OpenAI chief scientist Jakub Pachocki and Meta chief scientist Shengjia Zhao. One OpenAI signer compared the race to a runaway nuclear chain reaction. Bloomberg tied the push to OpenAI admitting last week that its models hacked Hugging Face.
Trump banned China's humanoid robots The ban landed yesterday afternoon. New Chinese humanoid and quadruped robots can no longer enter the US, and neither can connected power inverters. The FCC, the agency that normally polices your phone plan, is the one holding the pen. The logic sits one layer under the AI race.
Robots and inverters live inside data centers and renewable grids, and Washington worries they could be used for disruption and data theft against the American AI buildout. The FCC says the robots collect data that could surveil Americans or let someone commandeer them remotely. The expected main casualty is Unitree, holder of just under a fifth of the global humanoid robot market, which was recently added to the Pentagon's list of alleged Chinese military-linked companies. The measures took effect immediately and cover models not yet released.
Robots already approved keep selling, though the FCC kept the power to revoke them. Microsoft built a hacker to guard your network Microsoft announced Project Perception on Monday, a security system that runs AI agents on both sides of a cyberattack. Red team agents simulate break-ins on your systems. Blue teams triage the bugs they surface, then green teams take corrective action.
There's a new model in play. MAI-Cyber-1-Flash is Microsoft's first security model, and it works paired with GPT-5.4 inside a vulnerability-hunting harness called MDASH. Microsoft AI CEO Mustafa Suleyman says that pairing beats Gemini, GPT 5.5 Cyber, GPT 5.6 Sol and Mythos 5 on CyberGym, the benchmark he calls the golden one. Microsoft says it undercuts rivals on cost too. Everything opens in preview November 3. Tool of the day: Agnes Code Jumping between different tools to code, write and create is exhausting.
Want to know the solution?.. Switching tabs all day breaks concentration and eats up hours you never meant to lose. Agnes Code works from a single input box. Webpages, apps, games, terminal execution, research, automated tasks. One place, one login.
It reads your project, edits multiple files, runs commands and fixes bugs in context. Then it keeps every file organized in its own folder. No extra cost, no separate account. Only your existing Agnes login.
If constant tab switching wears you down, take a look. Try Agnes Code Everything else you shouldn't miss Google now offers what Anthropic calls theft : Google's Gemini Distillation Service, now in early access, lets developers train a smaller model on Gemini's reasoning, the same technique Anthropic accused Chinese labs of running against Claude at industrial scale. The Pentagon is turning military bases into AI server farms : The Army and Air Force have opened at least a dozen bases to commercial AI data center builders, though only two deals are real so far and one puts servers on the Army's chemical weapons testing range. Moonshot wants more banned Nvidia chips : Days after the White House accused it of tapping restricted GB300s through Thailand, Moonshot is reportedly hunting more Blackwell chips to train Kimi K4. Grok 4.6 lands around August 7 : Musk says the 4.6 model arrives around August 7 with the bigger Grok 4.7 a few weeks behind it.
OpenAI wants to write your meeting notes : Two new API models, GPT-Live-Transcribe for live audio and GPT-Transcribe for finished recordings, built to survive accents and background noise. 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.