Anthropic's Accidental Transparency: 512,000 Lines of Claude Code Exposed

Anthropic's accidental exposure of 512,000+ lines of Claude Code source code via a misconfigured npm file revealed undisclosed features including 'Undercover Mode' (preventing Claude from revealing internal codenames), user emotion detection, and an unreleased proactive assistant called KAIROS. The leak — the first time a leading AI company's full internal agent architecture has been publicly exposed — forces a broader reckoning with the transparency gap between what AI companies tell users and what they actually build. The episode also covers PrismML's unverified but potentially significant one-bit model claims, Oracle's massive layoffs, Salesforce's AI-heavy Slack overhaul, and a Stanford vision study with thin sourcing that nonetheless raises important quest ...