https://blog.thomas-bray.com/categories/ai/Recent content in AI on Rabid CuriosityHugoen-usThomas BrayThu, 23 Apr 2026 06:30:00 -0500https://blog.thomas-bray.com/posts/when-the-vault-has-a-back-door/Thu, 23 Apr 2026 06:30:00 -0500https://blog.thomas-bray.com/posts/when-the-vault-has-a-back-door/<p>Anthropic&rsquo;s Claude Mythos leak is not a surprise. It was an inevitability. And it teaches us something important about how power actually works in the information age — and why the entire edifice of information security is built on sand.</p> <h2 id="the-paternalism-model">The Paternalism Model</h2> <p>Anthropic designed Mythos as a cyber-capable AI so powerful it could find and exploit vulnerabilities in every major operating system and browser. Their solution? Keep it locked up. Only trusted partners — AWS, Apple, Google, Microsoft, the usual suspects — would get access through the Project Glasswing initiative.</p>