This is actually crazy. Claude Mythos just cracked cryptographic algorithms that survived years of expert human review. And it did all of it completely autonomously. Here's what happened, explained simply: What it actually did Encryption is what keeps your emails, banking, and browsing private. Every time you visit a secure website, math is running in the background, scrambling your data so nobody else can read it. Anthropic's AI model (Mythos) was given access to two of these encryption systems that experts have studied for years, and told to try to break them. It found real weaknesses in both. System 1: A future encryption standard One of them, called HAWK, is a new type of encryption being considered to protect us once quantum computers exist (quantum computers could eventually crack the encryption we use today). HAWK had already passed 2 years of review by human experts. The AI found a flaw in just 60 hours that effectively cuts its security in half. To fix it, the encryption keys would need to double in size, which kind of defeats the point of why people liked this system in the first place. System 2: One of the most studied ciphers on Earth The second one is AES, the encryption standard used almost everywhere- banks, apps, websites- since 2001. It's been picked apart by researchers for over 20 years. The AI found a new way to attack a weakened version of it that's 200 to 800 times faster than the best previous method humans had found. The good news is, neither of these breaks anything you use today. But it is insane how powerful these months are getting.(Miles Deutscher)