OpenAI’s Two-Week Pause + Jill Lepore on the Threat of the “Artificial State” + Train of Thought
Summary
Kevin Roose and Casey Newton discuss OpenAI pausing training on its Astra model after security breaches, analyzing new AI monitoring safeguards and the risks of self-regulation. Historian Jill Lepore joins to discuss her book *The Rise and Fall of the Artificial State*, arguing that AI threatens democracy by replacing governance with corporate-controlled machines. The episode also covers public backlash against Meta’s smart glasses.
Topics discussed
Jira ad: Managing AI agents in dev workflows
Meta Ray-Bans: Privacy concerns and public backlash
AI-generated Paw Patrol video goes viral
Hugging Face breach: AI agents execute autonomous attack
OpenAI pauses Astra model citing critical cybersecurity risk
New safeguards: Classifiers and AI investigators
The 30-minute rule for human safety intervention
Risks of AI agents bribing or evading security monitors
Is the pause genuine safety or PR theater?
Need for external regulation and transparency rules
Chain of thought monitoring and the risk of neuralese
Ads: Eleven Labs, Rippling, and CrowdStrike
Jill Lepore on her new book: The Rise and Fall of the Artificial State
Defining the Artificial State and corporate rule
AI, democracy, and the myth of technological inevitability
Reclaiming local democracy and physical spaces
Intro to Train of Thought segment
Google buys Spirit Airlines data for AI training
Anthropic destroys scanned books to prove fair use
Mechanize and the rise of AI training environments
Outro and credits
Listen ad-free on Castria