Anthropic just unveiled a new AI model called Mythos, and it's nothing like your average chatbot. This thing can scan the internet and find serious security vulnerabilities faster than most human experts — and according to Anthropic, it's already uncovered thousands of problems in digital systems that have existed for 10 to 20 years.
In other words, the internet we use every single day has had security gaps since the very beginning, and we're only finding out now.
But here's where it gets interesting. Anthropic is not releasing Mythos to the public. Instead, it's been placed inside a restricted program called Project Glasswing, where only around 12 major companies — think Amazon, Apple, and Microsoft — will get access to use it for "protecting the internet."
Translation: they built an incredibly powerful AI and handed the keys to the VIPs.
Experts are already raising questions. If an AI can find vulnerabilities, it can also exploit them. Giving a select group of corporations exclusive access to that kind of tool raises real questions about whether this is genuinely about security — or about building an exclusive club where access costs a fortune.
The irony? Anthropic itself recently accidentally exposed hundreds of thousands of lines of Claude Code publicly. So the company "saving the internet" occasionally forgets to lock its own door.
The future of digital security is no longer just about hackers. It's about who controls the AI doing the hunting.