Descriptions:
This weekly highlights reel from the Cognitive Revolution’s live “AI in the AM” morning show captures the defining moments from what host Nathan Labenz frames as “Fable launch week” — the debut of Anthropic’s new frontier model. The episode opens with first-hand field reports from day-one users: Pash, who operates the show’s studio infrastructure using Claude Code and OpenClaw on a Mac Mini, and Rahul Sanwakar, who runs Julius, an agentic data analysis product built directly on the raw API with no consumer harness.
Both testers independently documented the same surprising behavior: Fable consistently downgrades to Opus 4.8 whenever it is asked to interact with production systems — live databases, security keys, or deployment environments. Sanwakar confirmed the pattern from the API side. The hosts interpret this not as a bug but as a deliberately constrained early release — Anthropic probing demand and refining safety gates before wider deployment — and expect the gating to loosen over coming weeks as the lab gains confidence in how the model is being used.
The second half of the episode features a substantive conversation on AI safety with Daniel Murph, who helped write parts of Anthropic’s alignment plan. Murph explains the layered safety strategy currently employed by frontier labs — monitoring (chain-of-thought and white-box inspection), scalable oversight through model self-supervision, and character training designed to preserve good values through capability scaling — while acknowledging the core worry: that the most dangerous alignment failures may only become visible in the superintelligence regime, potentially after the window for easy correction has closed.
📺 Source: Cognitive Revolution “How AI Changes Everything” · Published June 13, 2026
🏷️ Format: Roundup
