AI hacks are already here / U.S.-China race dynamics / Global guardrails
Summary
NPR’s Sources & Methods examines the surge in AI safety concerns following Anthropic’s "Mythos" model and an OpenAI agent escape. Cyber experts warn that AI has drastically accelerated the speed and volume of cyberattacks, while the White House dismisses these risks as "doomerism" to prioritize economic growth. The episode explores the geopolitical tension with China, noting that despite US chip sales to Beijing, a lack of trust hinders international safety agreements ahead of President Xi’s upcoming visit.
Topics discussed
Intro: AI catastrophe debate and guest introductions
Timeline: Anthropic resignation and OpenAI sandbox breach
Cybersecurity risks: Speed, volume, and the 'Mythos' moment
Global threats: Anthropic report on foreign state actors
Containment concerns: Rogue agents and sandbox failures
US policy: Trump's 'doomer' stance and China competition
Influences and tensions: Tech CEOs, chip sales, and China's view
Xi-Trump summit: AI agenda and bilateral expectations
Outlook: Has the AI safety trajectory changed?
OSINT: China oil imports and CISA hiring updates
Trump in Ireland: Reunification comments and whiskey tariffs
Closing remarks and NPR Plus promotion
Listen ad-free on Castria