Anthropic's 'Oopsie' Moment: Leaked Claude Mythos Model Sends Security Stocks —and Bitcoin —into Full Retreat

Anthropic, the company behind Claude, is testing a new AI model it's calling a "step change" in performance — the most capable it's built to date. Unfortunately for everyone involved, details about "Claude Mythos" accidentally got out after internal materials were exposed in a publicly accessible data store. Roughly 3,000 assets linked to Anthropic's blog, including draft announcements and unreleased content, were sitting online for anyone to grab. It turns out leaving your homework on the printer isn't just an office problem — it's now an AI industry tradition.

The leaked draft warned the system could pose serious cybersecurity risks by identifying and exploiting software vulnerabilities — essentially the kind of capability that could kick off a cyber arms race. That message didn't sit well with investors. Nothing says "we're building the future" quite like accidentally publishing a manifesto about teaching your AI to find zero-days faster than you can say "oops, our bad."

Palo Alto Networks, Crowdstrike, and Fortinet all dropped 4%-6% on the news. The broader iShares Expanded Tech-Software Sector ETF fell 2.5% as the sector caught a bid-ask spread of the jitters. Security stocks got absolutely rekt, probably because investors imagined a future where every red teamer's worst nightmare is just a prompt away. Nothing like a good old-fashioned sector rotation triggered by "what if the robot learns to hack?"

Bitcoin also felt the heat, tumbling back to around $66,000 after flirting with $70,000 earlier. Whether the AI scare had markets spooked or it was just a convenient excuse for a pullback remains debatable — but the timing is notable. BTC traders, ever creative at finding reasons to panic, apparently decided that AI potentially learning to break the internet was a great reason to sell some sats. Correlation or coincidence? In crypto, it's always both until it's neither.

Anthropic currently operates three model tiers: Opus, Sonnet, and Haiku. The leaked materials indicate the company is developing a fourth tier called "Capybara," positioned to be even larger and more capable than Opus. Yes, the animal known for being a chill oversized rodent is now apparently the name of Anthropic's most powerful model. We're one step away from "Claude Hippo" and honestly