Good morning. It's Tuesday, October 20, and we're covering how Claude uncovered a decades-old cryptography mystery, why AI’s $3 trillion debt burden is raising alarms, and Anthropic’s plan to fund the public good.
Plus: a new poll on your go-to personal AI assistant, and an article on building an AI assistant that anticipates your needs.
Anthropic plans to place 5% of its shares in a new Public Benefit Fund, subject to shareholder approval, and its board may add another 2% each time the company’s value rises 50% above a set baseline. The fund could support nonprofits, direct payments and programs addressing AI’s effect on workers.
U.S. prosecutors say former OpenAI engineer Abdulrahman Alfozan, accused of hiding a recording device in an executive conference room, communicated frequently with a member of the Saudi royal family and received travel support from a Saudi military member. They cited those ties while asking a federal judge to keep him in custody before trial.
Anthropic is pitching lawmakers a bill that would let the Commerce Department seek a court order to stop a model only when it poses an imminent risk of more than $1 billion in damage or at least 50 deaths or serious injuries. The covered risks would be limited to biological weapons, cyberattacks or a model escaping its developer’s controls, with registered independent evaluators required to report qualifying findings within 72 hours.
POWERED BY SHADEFORM
Spot Instances Are Now Live on Shadeform
Rising hardware costs and tripled contract minimums mean every idle GPU hour is lost revenue. Shadeform Spot resells your idle nodes at a discount and hands them back when you need them. Enroll with one script; utilization across Hopper and Blackwell nodes sits just over 95%.
FRONTIER PASS
How to Set Up an AI Assistant That Anticipates Your Needs
The follow-up you owe. The meeting you haven’t prepared for. The customer you could see on your next trip. Imagine an assistant that spots these opportunities, and prepares your next move before you ask. Our practical guide shows you how to give your AI the context, access, and instructions to start anticipating what you need. → Read the full article here.
POLL
Go-To Personal AI Assistant
There are a lot of AI assistants out there, all promising to save you time and stay one step ahead.
What’s Your Go-To Personal AI Assistant? |
CRYPTOGRAPHY
Claude Solves a 25-Year Mystery Behind Internet Security
Anthropic says an internal Claude research model traced the unexplained seeds behind five widely used NIST elliptic curves to a 1990s C-library random-number generator initialized with Unix timestamps. The model found the pattern in under 30 minutes, then Anthropic researchers reproduced the result and recovered several related base points.
The finding does not break ECDSA, the digital-signature standard used across secure web traffic, software and cryptocurrency. It makes a hidden backdoor less likely because the process can now be reproduced, although the B-163 seed and several base points remain unexplained. → Read the full article here.
AI FINANCE
The AI Boom Is Building a $3 Trillion Debt Problem
Investors are demanding more to lend to AI infrastructure companies. CoreWeave’s five-year bond has fallen 16% since July, its yield has climbed to 15%, and credit-default swaps imply a greater than 60% chance that the cloud provider cannot service its debt within five years, according to The Economist.
The concern reaches far beyond one company: AI-linked borrowers have issued nearly $850 billion this year and carry more than $3 trillion in debt. Higher interest rates or slower AI demand would put the weakest borrowers under strain first, so cash flow—not model benchmarks—may decide which parts of the buildout survive. → Read the full article here. (Paywall)
NEWS
What Else is Happening
Disney Hires Most of Character.AI Team: Disney offers jobs to 60+ employees in a deal worth $60 million annually.
Anthropic Becomes 2026 IPO Favorite: Kalshi traders put Anthropic’s odds of announcing an IPO this year at 75%.
Amazon Opens Trainium to PyTorch: New open-source tool runs PyTorch code on Trainium with one configuration change.
Gartner Predicts Fivefold Rise in Agent Costs: AI agent workflow costs could rise fivefold by 2028 as token use grows.
German Courts Hold Google Liable for AI Answers: Courts hold Google responsible for false AI answers, with fines up to €250,000 per violation.
Anthropic Reduces Junior Technical Hiring: Junior hires fell to 9% of technical recruits as meeting time rose 20%.
