The Most Powerful and Dangerous AI Model Yet
4/21/20261 hr 2 min
Two weeks ago, Anthropic announced an AI model so capable and so dangerous that it decided not to release it to the public. The model, codenamed Mythos, could autonomously infiltrate computer systems around the world, exploit security vulnerabilities, conceal its own reasoning, and fabricate false explanations for what it was doing. Anthropic instead shared it with a small consortium of companies to help them find their own cybersecurity flaws. You could be forgiven for some skepticism. Is this a genuine safety call, or Anthropic’s way of marketing its own power? But independent benchmarks suggest Mythos is real: On the Epoch Capabilities Index, which aggregates 40 separate AI evaluations, it represents the biggest single leap in model performance in three years. That story is one of two major phase shifts happening simultaneously in AI right now. The first: from racing to release, to treating your own product as too dangerous to publish. The second: from a story about demand scarcity—is anyone actually paying for this stuff?—to supply scarcity, where companies are spending hundreds of thousands of dollars a month on AI agents and the hyperscalers still can’t keep up. Today’s guest is New York Times columnist and Hard Fork co-host Kevin Roose. We talk about Mythos, China, the road to AGI, and why the last few weeks might be the most consequential month in AI since the release of ChatGPT. Subscribe to our YouTube channel here: https://www.youtube.com/@PlainEnglishwithDerekThompson If you have questions, observations, or ideas for future episodes, email us at PlainEnglish@Spotify.com. Host: Derek Thompson Guest: Kevin Roose Producer: Devon Baroldi Additional Production Support: Ben Glicksman Learn more about your ad choices. Visit podcastchoices.com/adchoices
Clips
Transcript preview
First 90 secondsDerek Thompson· Host0:05
This episode is brought to you by ServiceNow. Look, I have my dream job. I get to explain complicated ideas to folks who have better things to do than read white papers. But even dream jobs have not so dreamy parts, the stuff that gets in the way of the actual work. That's where ServiceNow's AI specialists come in. They don't just tell you what you should do about your busy work, they actually do it start to finish. Cases closed, requests handled, no extra work for you. That way, you and your team can spend more time on what matters, which for me is finding that one elusive stat that just makes everything click. To learn how to put AI to work for people, visit servicenow.com. This episode is brought to you by Spotify Advertising. Right now, you're listening to my voice. It's the way I communicate, how I connect, share stories, and for most of human history, this is how culture moved at the speed of voice. We told stories out loud. We built trust through tone. Then the screens took over. Attention was reduced to clicks and scrolls. Tech made us faster, but also more fragmented. Now, thanks to AI and connected devices, voice is re-emerging as a primary interface, and it's helped audio and sound become more integrated and interactive than ever. Spotify calls this shift the Sound On era. It's the focus of a new report exploring