Anthropic built a monster. Claude Mythos is too good at breaking things, so they hid it from the public. Access is restricted to a tiny circle of security experts and massive orgs. Smart move. Dangerous, but smart.
Now we see what it can do. The Wall Street Journal got the scoop. Security researchers at Calif, a Palo Alto firm, used Mythos. They cracked Apple’s macOS.
“First public macOS kernel memory corruption exploit on an Apple M5 chip”
That is what Calif called it on Thursday. No fluff. Just the facts. The exploit turns an unprivileged user into a god. Total control of the device. How? Two vulnerabilities. Several tricks. But Mythos did the heavy lifting. It found the bugs. It helped write the code to exploit them.
It’s fast. Too fast.
“Once it has learned how to attack one problem, it guesses the answer for nearly any similar one.”
Mythos knows the class of bug. It applies the pattern. It generalizes. The researchers didn’t search; the AI predicted where the weak spot was.
Did Apple fix it?
Maybe. macOS Tahoe 26.6 came out Monday. The release notes mention a fix submitted by Calif and Anthropic. Calif’s name shows up in other reports too. MacRumors spotted the connection immediately. But Calif’s blog post leaves things muddy. They met Apple early this week. Maybe it is fixed. Maybe it is still wide open.
Details are withheld until Apple patches everything. Of course. Standard procedure. An Apple spokesperson told the WSJ that security is a priority. Standard corporate speak. Canned and clean.
Mythos isn’t done looking. It is just starting to hunt. 🕷️
