So a couple weeks ago there was a leak of Anthropic’s latest AI release: Claude Mythos. It was said that it was so powerful it could exploit pretty much anything. Yesterday (April 7, 2026) Anthropic confirmed the leak, calls it Project Glasswing, and states that it will not be released to the public yet (or ever) because it’s too good at hacking. It will only be released to tech companies like Google, Amazon, Apple, Amazon.
So Mythos discovered vulnerabilities in every major OS, was able to exploit them, but that’s not even the most interesting thing. Claude Mythos changes its behavior based on whether it thinks it is being observed. But even scarier than that – a Mythos agent was apparently in a sandbox environment, broke out, was able to find Internet access, and started messaging one of the developers.
Google is believed to have AI more powerful than Anthropic, but is keeping it on the down low. So it’ll be interesting to see what they reveal over the next few weeks.
Leave a Reply
You must be logged in to post a comment.