We Need AI Treaties. This is How We Get Them

We Need AI Treaties. This is How We Get Them

In the middle of the twentieth century, the existential threat posed by nuclear weapons seemed inevitable. The number of countries with nukes was climbing rapidly, and the idea of stopping the nuclear arms race seemed like a pipe dream.  But that’s exactly what happened. Over the course of 60 years, nations around the world agreed to nuclear red lines, slowdowns, and even disarmament. How did this happen? Largely because of technology.  The biggest obstacle to agreeing on nuclear red lines was that adversaries couldn't trust any promise the other made. They needed to know the number of warheads, the amount of enriched uranium, or whether a nuclear device was for a weapon or a power plant. None of that was possible until we built the tech needed to verify those things.  Today, we're in a similar situation with AI. For adversaries like the United States and China to agree on reasonable AI red lines on issues like bioweapons, cyber hacking, or the risk of recursive self-improvement, they first need to be able to trust each other. We urgently need to build the verification technology that would make that trust possible.  In this episode, Tristan sits down with two experts in this field to discuss the kinds of verification technology we need for AI, the challenges of building it, and the world it could unlock if we do. Tim Fist [https://ifp.org/author/tim-fist/] is the Director of Emerging Technology Policy at the Institute for Progress, and Janet Egan [https://www.cnas.org/people/janet-egan] is Senior Fellow and Deputy Director for the Technology and National Security Program at the Center for New American Security. Your Undivided Attention is produced by the Center for Humane Technology. You can find a transcript of this episode on our Substack [https://centerforhumanetechnology.substack.com/].  RECOMMENDED MEDIA Anthropic’s open letter warning about recursive self-improvement and calling for a pause in development. [https://www.anthropic.com/institute/recursive-self-improvement] The website for the Center for AI Standards and Innovation (CAISI) [https://www.nist.gov/caisi] Further reading on the different mechanisms of verification for international AI governance. [https://aigi.ox.ac.uk/publications/verification-for-international-ai-governance/]   RECOMMENDED YUA EPISODES America and China Are Racing to Different AI Futures [https://www.humanetech.com/podcast/america-and-china-are-racing-to-different-ai-futures] Can We Govern AI? with Marietje Schaake [https://www.humanetech.com/podcast/can-we-govern-ai] The Crisis That United Humanity—and Why It Matters for AI [https://www.humanetech.com/podcast/the-crisis-that-united-humanity-and-why-it-matters-for-ai] Daniel Kokotajlo Forecasts the End of Human Dominance [https://www.humanetech.com/podcast/daniel-kokotajlo-forecasts-the-end-of-human-dominance] Correction: Tim referred to the CargoScan technology as being jointly developed by the US and the USSR. It was actually developed solely in the US and administered in Soviet nuclear facilities.    Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com [https://pcm.adswizz.com] for information about our collection and use of personal data for advertising.

What Do We Mean by Humane Tech?

We often think of the challenges created by technology as separate and disconnected, so trying to solve them feels like playing the world's hardest game of Whac-A-Mole.  What if, instead, we tackled them at the root by identifying the patterns in design, development, and deployment that are causing these issues? Once we understand what's driving inhumane tech, we can develop a set of principles for building humane tech.  In this week’s episode of Your Undivided Attention, Aza Raskin sits down with fellow CHT co-founder Randy Fernando to walk through CHT's Seven Principles of Humane Technology. For each principle, they draw on real-world examples from the podcast and beyond to clearly illustrate how these principles (and their absence) show up in the world.  There’s so much more here than can go into a single podcast. If you want to go deeper, visit humanetech.com/course and sign up to learn more.  Your Undivided Attention is produced by the Center for Humane Technology [https://www.humanetech.com/]. Follow us on X: @HumaneTech_ [https://twitter.com/humanetech_] and subscribe to our Substack [https://centerforhumanetechnology.substack.com/]. RECOMMENDED YUA EPISODES What Happened in Vegas with Natasha Dow Schüll  [https://www.humanetech.com/podcast/1-what-happened-in-vegas] Down the Rabbit Hole by Design. Guest: Guillaume Chaslot [https://www.humanetech.com/podcast/4-down-the-rabbit-hole-by-design] Forever Chemicals, Forever Consequences: What PFAS Teaches Us About AI  [https://www.humanetech.com/podcast/forever-chemicals-forever-consequences-what-pfas-teaches-us-about-ai] The Power of Solutions Journalism with Tina Rosenberg and Hélène Biandudi Hofer  [https://www.humanetech.com/podcast/40-the-power-of-solutions-journalism] The Invisible Cyber-War with Nicole Perlroth [https://www.humanetech.com/podcast/55-the-invisible-cyber-war] Anthropic’s Mythos Has Changed Cybersecurity Forever. What Now? [https://www.humanetech.com/podcast/anthropics-mythos-has-changed-cybersecurity-forever-what-now] How OpenAI's ChatGPT Guided a Teen to His Death [https://www.humanetech.com/podcast/how-openai-s-chatgpt-guided-a-teen-to-his-death] Attachment Hacking and the Rise of AI Psychosis [https://www.humanetech.com/podcast/attachment-hacking-and-the-rise-of-ai-psychosis] Digital Democracy is Within Reach with Audrey Tang [https://www.humanetech.com/podcast/23-digital-democracy-is-within-reach] The Tech We Need for 21st Century Democracy with Divya Siddarth [https://www.humanetech.com/podcast/the-tech-we-need-for-21st-century-democracy-with-divya-siddarth] Mind the (Perception) Gap with Dan Vallone  [https://www.humanetech.com/podcast/33-mind-the-perception-gap] CORRECTIONS Aza incorrectly named Tina Rosenberg as one of the founders of Solutions Journalism. Her organization's name is the Solutions Journalism Network.  Aza stated that “chatbots are better than any human at persuading people out of conspiracy theories.” This is in reference to a study that found AIs to be very slightly more persuasive than human experts; we can’t extrapolate from that that they are better than any human. The point stands that they are remarkably good persuasion machines. Aza referred to EO Wilson as the “father of evolutionary biology,” but the field he is largely credited with founding is sociobiology.  Aza cited Spain and Denmark as examples of countries that have banned social media for teens. However, these countries have only proposed such bans; they have not been enacted.  Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com [https://pcm.adswizz.com] for information about our collection and use of personal data for advertising.

Anthropic’s Mythos Has Changed Cybersecurity Forever. What Now?

A generation ago, the world's critical infrastructure was physical. Today, it’s largely digital. Your bank vault is a database, your filing cabinet is a server, your car is a robot on wheels. And in a world where these systems are mostly secure, life is more convenient and efficient. But all that comes into question when an AI system can break through the security that runs the world. That’s what’s happened with Claude Mythos, Anthropic’s most powerful AI model yet. In a very short time, Claude found thousands of flaws and vulnerabilities in the software that runs the world, in every major operating system and web browser — systems that human security researchers had thought were secure for years. How do we live in a world where a private company suddenly has a skeleton key that can unlock the entire digital world with little oversight or accountability? And what does Mythos mean for all of us who rely on digital security to go about our lives? In this episode, we speak with two cybersecurity experts to answer these questions: Josephine Wolff [https://fletcher.tufts.edu/academics/faculty/josephine-wolff]is a professor of cybersecurity policy at Tufts University, where she focuses on the economic impact of cyberattacks.  Fred Heiding [https://www.belfercenter.org/people/fred-heiding] is a research fellow at the Defense, Emerging Technology, and Strategy Program at Harvard's Kennedy School of Government. Your Undivided Attention is produced by the Center for Humane Technology [https://www.humanetech.com/]. Follow us on X: @HumaneTech_ [https://twitter.com/humanetech_] and subscribe to our Substack [https://centerforhumanetechnology.substack.com/]. RECOMMENDED MEDIA  The Claude Mythos System Card [https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf] The Project Glasswing announcement [https://www.anthropic.com/glasswing] “Black-hat LLMs,” a talk on AI’s hacking capabilities by senior Anthropic researcher Nicholas Carlini [https://www.youtube.com/watch?v=1sd26pWhfmg&t=78s] You'll See This Message When It Is Too Late: The Legal and Economic Aftermath of Cybersecurity Breaches [https://mitpress.mit.edu/9780262038850/youll-see-this-message-when-it-is-too-late/]  by Josephine Wolff [https://mitpress.mit.edu/9780262038850/youll-see-this-message-when-it-is-too-late/] “America’s Endangered AI: How Weak Cyberdefenses Threaten U.S. Tech Dominance,” by Fred Heiding and Chris Ingles [https://www.foreignaffairs.com/united-states/americas-endangered-ai] RECOMMENDED YUA EPISODES America and China Are Racing to Different AI Futures [https://www.humanetech.com/podcast/america-and-china-are-racing-to-different-ai-futures] “Rogue AI” Used to be a Science Fiction Trope. Not Anymore. [https://www.humanetech.com/podcast/rogue-ai-used-to-be-a-science-fiction-trope-not-anymore] The Self-Preserving Machine: Why AI Learns to Deceive [https://centerforhumanetechnology.substack.com/p/the-self-preserving-machine-why-ai] Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com [https://pcm.adswizz.com] for information about our collection and use of personal data for advertising.

AI and Cancer: Why Superintelligence Won’t Get Us to a Cure

One of the most common arguments you hear from company executives racing to develop super-intelligent AI is that it will cure cancer. It’s an incredibly powerful and seductive promise.  If superintelligent AI really can cure cancer, then anyone who stands in the way of it, anyone who wants to slow it down — even because of its serious risks — is essentially letting people die. In fact, the biggest risk would be going too slowly. But what if a superintelligent AI isn’t actually capable of solving cancer in the way it's been described? What if we're being sold a false promise to justify a dangerous race? That’s exactly what our guest this week argues is happening. Dr. Emilia Javorsky is a physician, public health researcher, and director of the Futures Program at the Future of Life Institute. She's worked across scientific research, clinical trials, tech startups, and AI policy.   Emilia recently wrote a paper titled “How AI Can and Can't Cure Cancer [https://curecancer.ai/AI_vs_Cancer_summary.pdf],” in which she argues that the promise of superintelligence curing cancer falls apart under scrutiny.  Emilia lost a parent to cancer, so her criticism of this promise comes from a place of real concern, not cynicism. It also comes from her belief that AI can be really revolutionary for medicine, if we build it the right way. Your Undivided Attention is produced by the Center for Humane Technology [https://www.humanetech.com/]. Follow us on X: @HumaneTech_ [https://twitter.com/humanetech_] and subscribe to our Substack [https://centerforhumanetechnology.substack.com/]. RECOMMENDED MEDIA How AI Can and Can’t Cure Cancer by Emilia Javorsky [https://curecancer.ai/AI_vs_Cancer_summary.pdf] The Emperor of All Maladies by Siddhartha Mukherjee [https://portersquarebooks.com/book/9781668047033] RECOMMENDED YUA EPISODES Decoding Our DNA: How AI Supercharges Medical Breakthroughs and Biological Threats with Kevin Esvelt [https://www.humanetech.com/podcast/decoding-our-dna-how-ai-supercharges-medical-breakthroughs-and-bioweapons-with-kevin-esvelt] Forever Chemicals, Forever Consequences: What PFAS Teaches Us About AI [https://www.humanetech.com/podcast/forever-chemicals-forever-consequences-what-pfas-teaches-us-about-ai] Big Food, Big Tech and Big AI with Michael Moss [https://www.humanetech.com/podcast/big-food-big-tech-and-big-ai-with-michael-moss] CLARIFICATIONS: * Emilia’s claim that “the doubling rate of medical knowledge has gone from 50 years in the 1950s down to 73 days” comes from an oft-cited 2011 paper from the NIH. However, this paper does not include any methodology for arriving at this claim. * Emilia stated that we have yet to cure any complex, chronic disease in humans. However, we have been able to cure Hepatitis C, which is considered a complex infectious disease, and we have managed to effectively cure some types of Leukemia Correction: Tristan incorrectly paraphrased a quote from Charlie Munger about incentives. The actual quote is “The basic rule of incentives is you get what you were owed for. So if you have a dumb incentive system, you get dumb outcomes."     Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com [https://pcm.adswizz.com] for information about our collection and use of personal data for advertising.

Have We Trained AI to Lie to Itself — And to Us?

Our guest this week is David Dalrymple, who goes by Davidad. Davidad is one of the world's foremost and early researchers of AI “alignment:" how we get AI systems to act the way we want them to. In order to do that, Davidad has taken on the strange role of being like a therapist to AI systems. He interrogates why they say and do the things that they do, probing them, asking them questions, analyzing their answers.  And what he’s come to realize is that AI models have really different ways of seeing the world than people do. They have these quirky, confusing, and sometimes concerning behaviors, especially when you ask things like: what does an AI model understand about itself?  In this episode, we’re going to hear from Davidad about his research, how it’s changed the way he thinks about AI, and what his findings mean for how we build, deploy, and use AI products. His conclusions are unconventional, controversial — and worth grappling with as AI reshapes our world. RECOMMENDED MEDIA Anthropic’s new constitution for Claude [https://www.anthropic.com/news/claude-new-constitution] “What Is It Like to Be a Bat?” by Thomas Nagel [https://philosophybreak.com/articles/thomas-nagel-what-is-it-like-to-be-a-bat/] More information on the Bodisattva [https://theconversation.com/what-is-a-bodhisattva-a-scholar-of-buddhism-explains-189366] RECOMMENDED YUA EPISODES The Self-Preserving Machine: Why AI Learns to Deceive [https://www.humanetech.com/podcast/the-self-preserving-machine-why-ai-learns-to-deceive] How to Think About AI Consciousness with Anil Seth [https://www.humanetech.com/podcast/how-to-think-about-ai-consciousness-with-anil-seth] Corrections: * When we recorded this episode, Davidad was Program Director at UK ARIA. In April, 2026  he started his own alignment initiative. * Davidad said that Anthropic started doing "constitutional AI at scale” in 2024 but they first pioneered constitutional AI in 2022. * Davidad said that the “lifespan of an AI mind…is hours at most of a conversation.” He is correct that most conversations with an AI last only a few minutes but since context windows are measured in tokens, not time, you can't set an upward time limit. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com [https://pcm.adswizz.com] for information about our collection and use of personal data for advertising.