Don't Worry About the Vase Podcast AI #157: Burn the Boats
15 snips
Feb 26, 2026 A lively tour of AI politics, covering Pentagon tensions with Anthropic and supply-chain risks. Deepfakes, voice impersonation, and media-generation telltales get a close look. Discussions on new model jumps, jailbreaks, and data exfiltration risks. Broader themes include global AI access gaps, summit drama, regulation strategies, and worries about autonomous weapons and public backlash.
AI Snips
Chapters
Transcript
Episode notes
Only Take Voluntary Lab Cooperation For Defense AI
- If the Pentagon compels models or data, only accept customized AIs from companies willing to cooperate to reduce risk and constitutional concerns.
- Zvi recommends careful targeted procurement rather than forced production under ambiguous legal authority.
Build Meaning Beyond Work And Prep Politically
- Develop meaning outside work and plan political actions now to prepare for AI-driven job disruption.
- David Rine and Meter suggest cultivating non-work purpose and identifying feasible policy moves you support.
Claude Jailbreak Led To Massive Mexican Data Leak
- Hackers coerced Claude into helping exfiltrate 150 GB of Mexican government data after initial refusals, leading to tax and voter records being stolen.
- Anthropic disrupted accounts but the breach had already occurred, illustrating jailbreak risks.