July 28, 2026
Cracking jokes while AI cracks locks
Discovering Cryptographic Weaknesses with Claude
AI poked at our digital locks, and the comments went straight into panic, memes, and nitpicks
TLDR: Anthropic says its AI helped researchers find new ways to weaken two encryption methods, including one being considered for future use, though nothing in real-world products is affected right now. Commenters swung between doom-posting, model-comparison gossip, and blunt reminders that this is serious research—not proof your online banking is suddenly toast.
Anthropic dropped a very academic-sounding bombshell: its Claude Mythos Preview helped researchers find new weaknesses in two encryption systems, including one future-facing digital signature candidate and a watered-down version of the lockbox system used all over the internet today. Before anyone starts changing bank passwords in a cold sweat, the company says nothing currently used in everyday products is broken. But the community? Oh, they did what communities do best: immediately split into camps of panic, pedantry, and popcorn.
One crowd zoomed in on the most movie-worthy detail: an AI system found in 60 hours what a heavily reviewed candidate had survived for years. That led to some wonderfully dramatic reactions, including one commenter imagining the faces of national security officials everywhere. Another obsessed over the behind-the-scenes AI teamwork, pointing out that the big breakthrough came from two worker agents arguing with each other until one refused to give up. Yes, the hottest subplot here is basically "buddy cop AI, but for codebreaking."
But the thread also had a full-on reality-check squad. One commenter translated the news into plain English: this is not “the internet is broken,” it’s more like “researchers got better at attacking a test version of a lock.” Others raised eyebrows at Anthropic using Mythos Preview instead of a newer model, with whispers that the newer one may actually be less capable. And then came the driest, sharpest jab of all: if a human found a flaw, we’d call it research—so why are people suddenly acting like the sky falls when an AI does it? Classic comment-section energy: half alarm bells, half fact-checking, all drama.
Key Points
- •Anthropic says Claude Mythos Preview helped discover mathematical weaknesses in cryptographic algorithms, not just implementation bugs in cryptographic software.
- •The article’s first main result is an improved attack on HAWK, a third-round post-quantum digital signature candidate in NIST’s process.
- •Anthropic says the HAWK attack was improved in about 60 hours of work and effectively cut the scheme’s key strength in half.
- •The second main result is a new attack on a reduced-round version of AES that removes one attacker guess and speeds prior best attacks by 200–800×.
- •The article states that neither finding affects current production systems because HAWK is not deployed and the AES result applies to a weakened research variant, not standard AES.