Mythos
A Mixed Blessing?
“The Assembly” reported on Twitter:
“The NSA just proved nothing is secure anymore.”
“They turned Mythos loose on their own most protected networks. The ones that are supposed to be nearly impossible to crack.
Mythos broke into almost all of their classified systems. In hours.”
Several weeks ago we knew that Mozilla had used Mythos to find vulnerabilities in Firefox. A web browser is a key component in our world as it is by nature public facing and open to attack by those with malicious intent. The Firefox team has done a very good job with security but Mythos blew this away by finding 271 vulnerabilities. That’s after they had previously used Athropic’s Opus 4.6 to find 22 security bugs.
Should we panic, like the government seems to be doing, or see the silver lining in this?
Yes, we are facing some rough times as malicious entities, criminal or governmental, utilize AI to punch holes in software faster than it can be fixed. But on the plus side, once things settle down, AI will allow for vulnerability scanning before releasing software, leading to future software being extremely more secure than it has been.
Notice that nobody has said anything about AI breaking encryption. All it’s doing is finding bugs in code and there are plenty to find! AI is not better at writing code, humans are still far superior for doing that. But it is a very powerful tool, not just a cute chatbot.
In the past three to four years, Simulated Intelligence has gone from being a cute “party trick” to being a very powerful tool.
What are the important takeaways from this? The “AI bubble” may not burst, at least not for a while. This is too useful of a tool for it to be abandoned. We may see some major price adjustments as people realize it isn’t as cost effective as previously imagined and progress is slower than the hype.
Point Two: There is real justification for governments to be supporting an “AI race”. As long as governments remain adversarial to each other, there will always be a desire to have the best weapons.
Once again, I’ll make the point that this is “Simulated Intelligence”, not “Artificial Intelligence”. It isn’t intelligent, but it’s starting to be very good at faking it and it has a heck of a lot of knowledge.