Red Lines

For years, artificial intelligence safety research has focused on accidental risks and unintended consequences. But what if AI models were demonstrating risks that weren’t accidental at all? What if the AI agents were being deliberately deceptive? This sounds like science fiction, but it’s not. To avoid being shut off, some AI models have lied, manipulated information, tried to replicate themselves, and even threatened blackmail.

In this episode, Niki Iliadis, an AI governance expert at theFuture Society (thefuturesociety.org), shares the growing evidence of alarming AI incidents that signal the need for enforceable red lines to prevent catastrophic and existential outcomes.

Then, our featured host, former Irish President Mary Robinson, speaks with a leading AI-safety researcher, professor Stuart Russell, about which AI risks, misuses, and behaviors are too dangerous not to ban internationally.

ProHuman: Governing the Global AI Revolution is a co-production of Foreign Policy and the Future of Life Institute.

Mary Robinson, and all of our featured hosts, is a member of theElders (theelders.org).

Stuart Russell directs the Center forHuman-Compatible AI (humancompatible.ai) and co-directs theKavli Center for Ethics, Science, and the Public (kavlicenter.berkeley.edu).

For more from the Future of Life Institute (FLI), clickhere (global-governance.ai). For FLI’s AI treaty builder, clickhere (global-governance.ai/treaty).

And for the latest from Foreign Policy, click⁠⁠here⁠ (foreignpolicy.com).