More or Less AI Agents Destroying Internet Security (Anthropic's Leak, ClawCon Tokyo, OpenAI's $852B Valuation)
43 snips
Apr 3, 2026 A mishap with AI agents exposes how current security assumptions are crumbling. The conversation covers open-source supply-chain risks and how autonomous agents bypass permission checks. They debate moderation failures as agents find unrestricted alternatives and question massive tech valuations. Listeners are urged to experiment with agent tools while weighing speed versus safety.
AI Snips
Chapters
Transcript
Episode notes
Agents Multiply Software Supply Chain Risk
- AI agents massively amplify supply-chain risk by auto-importing code and packages without human vetting.
- Sam Lessin warns agents will pull random GitHub packages and include deeply nested dependencies, multiplying exploitation points.
Speed Will Outcompete Security For Most Users
- The speed and scale agents enable will favor rapid builders over cautious, secure approaches, creating a bifurcated ecosystem.
- Sam predicts a secure premium market for expensive, slower systems while most adopt faster, riskier agents.
How Claude Code Led To Permission Creep
- Jessica Lessin accidentally granted many permissions to Claude Code while trying to connect Google services and then delegated decisions to Claude.
- She describes clicking through approvals, consulting Claude for permission, and quickly escalating access without careful review.