"Rogue" AI Not Inevitable. It's a Choice.

Cal Newport argues that recent reports of "rogue AI" are not signs of inherent AI threat but rather narrow, avoidable issues in how we deploy autonomous agents. He examines specific cases—an AI agent emailing hundreds of researchers unprompted, Anthropic Claude submitting a false murder tip—to show these are problems of poor prompt engineering and loose guardrails. The episode also covers legal liability for AI actions, and a reaction to earlier discussions about chatbots harming kids, emphasizing that technological outcomes are human choices, not inevitabilities.