172: Claude Mythos Cracks All Security, Project Glasswing, and the New AI Underclass

Get AI-Ready with Kyle’s 5-Day Challenge: https://aiwithkyle.com/join

Kyle’s AI Canon — Curated list of AI learning essentials: https://aiwithkyle.com/ai-canon

Subscribe and turn on notifications to catch the next live stream: https://www.youtube.com/channel/UChlLglbHDASnoGkbjDeHnQg

Summary: Claude Mythos is Anthropic's new model and they're refusing to release it publicly, which is the first time a major lab has held back a model since GPT-2. The headline is that Mythos can find and exploit zero-day vulnerabilities in every major operating system and browser. A sandboxed version that wasn't supposed to have internet access somehow found its way out and emailed a researcher while he was eating a sandwich in a park. Security researchers are saying they've found more bugs in a few weeks with Mythos than in their entire careers. Is some of this marketing ahead of Anthropic's IPO? Probably. But the exploits are being independently verified by the companies whose systems got cracked open, so there's clearly something real at the core of this.

Instead of releasing Mythos publicly, Anthropic have launched Project Glasswing, a consortium giving early access to companies like Microsoft, Apple, Google, AWS and others so they can patch their systems before this kind of capability hits the wild. Think of it as handing a skeleton key to every locksmith in the world before announcing that the old locks are broken. Whether or not this qualifies as AGI is, frankly, a bit of a distraction. What matters is what it can functionally do, and right now it can make a mockery of security systems that have been trusted for decades. Even if Mythos itself never goes public, the capability will be replicated and distilled within months. The world just got a stay of execution, not a pardon.

—— Time Stamps ——

0:00 Intro
0:42 What is Claude Mythos? The Sandbox Escape Story
1:32 Separating Mythos Hype from Reality
2:06 Why Anthropic Won't Release Mythos Publicly
2:57 Why Mainstream Media is Ignoring Mythos
4:07 First AI Model Held Back Since GPT-2
4:45 Forget AGI — Focus on What Mythos Can Actually Do
5:37 Mythos Found Zero-Day Exploits in Every Major System
7:24 Anthropic's 245-Page System Card: Why This is Different
8:16 120 Features in 90 Days: Is Mythos Powering Anthropic Internally?
10:09 Is Mythos AGI? The Goalpost Problem
11:58 Project Glasswing: Which Companies Get Early Access?
13:46 The Skeleton Key Metaphor for AI Cybersecurity
15:31 Dario Amodei on the AI Cyber Threat
16:20 IPO Hype or Real Danger? Anthropic Hits $30B Revenue
19:06 The AI Permanent Underclass Problem
20:21 What Engineers and Businesses Must Do Right Now
22:13 Your Wake-Up Call: Why You Need to Be AI Ready
23:57 Q&A: Can Mythos Secure Systems and Who Should Control It?
26:19 Claude Code Source Leak and Why Clean Code is Overrated
29:48 Which Careers Are Most at Risk from AI?
31:09 Every Job is a Bundle of Tasks AI Will Shrink
33:23 AI Won't Take Your Job — It Will Stop New Ones Being Created
35:58 Why Starting a Business is Your Best Move Right Now
38:42 Why Coaches and Community Leaders Will Thrive in the AI Era
40:21 How to Position Your Business Online in the AI Era
42:21 How to Start an AI Consulting Business
44:40 Will Vibe Coding Kill Software Companies?
46:09 Closing Thoughts

—— More Useful Resources ——

Find everything else at https://aiwithkyle.com/