Since Anthropic announced in April that it was limiting the release of its new artificial intelligence model Claude Mythos due to its hacking capabilities, there have been growing fears about the power of cutting-edge AI.Last week some of the world's top AI CEOs urged US Congress to bring in controls on the sale of synthetic materials they believe could, with the help of AI, build dangerous bioweapons.At the same time Anthropic is warning there is a real risk of losing control of frontier AI, and that it might be time to contemplate a coordinated slowdown or a pause of its development.More InformationFeatured:Jeffrey Ladish, Palisade Research Connor Leahy,  Control AIHelen Toner, Centre for Security and Emerging Technology CreditsSteve Cannane, ReporterImage DetailsSince Anthropic announced in April that it was limiting the release of its new artificial intelligence model Claude Mythos due to its hacking capabilities, there have been growing fears about the power of cutting-edge AI.(Reuters: Dado Ruvic/Illustration)Program:More from AMTranscriptMelissa Clarke: Leading developer of artificial intelligence, Anthropic, is warning there's a real risk of losing control of frontier AI. It has already slowed the release of its model, Claude Mythos, over concerns about its hacking capabilities. Now, some of the world's top AI CEOs are urging the US Congress to bring in controls on the sale of synthetic materials they believe could, with the help of AI, build dangerous bioweapons. Four Corners reporter Steve Cannane has more.Steve Cannane: Jeffrey Ladish used to work inside the AI machine as part of the security team at Anthropic. He's now the director of Palisade Research in Berkeley, where he tests the latest AI models and the risks of humans losing control of them. He says the people he talks to inside the big AI companies are genuinely frightened about the forces they're unleashing.Jeffrey Ladish: I think the scariest thing I hear from people working inside of the companies is that they think they are on the way to recursive self-improvement. The point where AI agents are writing all of the code and doing all of the research for the next generation of AI agents. You have really smart AI making smarter AI making smarter AI and so on. And that's a runaway process, potentially. And they think that they're going to be able to do it. They're like, we are not that many years, maybe one, maybe two, maybe three years from that process happening. That terrifies me.Steve Cannane: It was these very concerns that were at the heart of a stark warning released by Anthropic via a blog post on Friday, Australian time. It suggested slowing down or pausing frontier AI development might be a good thing if there was a global mechanism to monitor and enforce it. Connor Leathy is the US director of Control AI. He believes the race to try and build machines smarter than humans should be halted.Connor Leathy: Almost everyone understands that if you build something vastly smarter than yourself and you don't know how to control it, which we currently do not know how to do, that's a very, very, very dangerous thing. We must create a global trust but verify regime where various countries such as the US, China and others can mutually verify that no one is attempting to build super intelligence and then commit to not doing so and including stopping, if necessary, rogue actors or private companies or others from attempting to build super intelligence.Steve Cannane: Helen Toner used to be on the board at OpenAI and is now the executive director at the Center for Security and Emerging Technology in DC. She says it would be a hard sell to pause development at the big US companies if others continue to charge ahead and develop AI.Helen Toner: What I do think would make sense would be to think really seriously about what it would look like to install brakes. We can't really hit the brakes right now because we don't really have them. But there's simple things like making sure any AI system that you're developing, do you have a way to turn it off if it's going rogue? Under what conditions might we actually agree that it's in everyone's interest to try and slow down or pause for a period?Melissa Clarke: Helen Toner from the Center for Security and Emerging Technology in Washington DC, ending that report by Steve Cannane. Appears In Tributes for renowned cancer specialist Richard ScolyerDuration: 20 minutes 4 seconds20m 6itemsIn this episode1 of 6Iran launches waves of missiles at IsraelDuration: 3 minutes 52 seconds3m 52s2 of 6Professor Richard Scolyer dies at 59Duration: 4 minutes 26 seconds4m 26s3 of 6King's birthday honours listDuration: 4 minutes 15 seconds4m 15s4 of 6Shark attacks prompts calls to make spearfishing saferDuration: 3 minutes 9 seconds3m 9s5 of 6Neurologists outcry over weedkiller reviewDuration: 3 minutes 26 seconds3m 26s6 of 6PlayingGrowing fears surround growth of AIDuration: 3 minutes 19 seconds3m 19s