Claude Mythos: Highlights from 244-page Release

The model, the mythos, the legend. We have a new best AI model, but not all of us. How good is it, what does it’s new offensive capabilities mean? Why does it’s 244 page report card remind me of Her, and why did the creator of Claude Code call it ‘terrifying’. 30+ highlights sourced by reading the paper in full, old-school, no AI summary.

https://80000hours.org/aiexplained


Check out my fast-growing (!) app, free to use, and code INSIDER15 for paid tiers: https://lmcouncil.ai

AI Insiders ($9!): https://www.patreon.com/AIExplained

Chapters:
00:00 - Introduction
00:56 - Internal Release + Availability
02:37 - General Capabilities
05:12 - Self-improvement?
06:15 - ‘Terrifying’ Landscape
11:07 - Safety Decision
13:22 - Coding
14:49 - Alignment, Awareness
19:52 - GUI for Agents/Claws + Hallucinations
21:34 - …Emotions?
25:29 - Her connection

244-page System Card: https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf

Project Glasswing: https://www.anthropic.com/glasswing
Zero-Day Details: https://red.anthropic.com/2026/mythos-preview/

Mythos ‘terrifying’: https://x.com/bcherny/status/2041605852382351666

New Yorker Altman/Amodei: https://archive.fo/20260406100412/https://www.newyorker.com/magazine/2026/04/13/sam-altman-may-control-our-future-can-he-be-trusted

Alignment Risk Update: https://www-cdn.anthropic.com/79c2d46d997783b9d2fb3241de43218158e5f25c.pdf

In a Park: https://x.com/sleepinyourhat/status/2041584808514744742

“Uhm” - https://x.com/thsottiaux/status/2041749947385815109


Non-hype Newsletter: https://signaltonoise.beehiiv.com/

Podcast: https://aiexplainedopodcast.buzzsprout.com/