How AI is working to make your iPhone and Mac nearly impervious to attacks

AI coding agents set to level the playing field between hackers and security engineers.

- JASON CROSS

Software security is a nightmare. You've probably noticed that every few weeks, there's an update for your Apple devices that patches dozens of vulnerabilities, and it never seems to stop. Modern software is so complex and so interconnected with other software that it's almost impossible to keep up with the threats.

The "attack surface" of any system is the total amount of potential areas of attack. It's all the code a hacker could find a hole in to compromise your device, program, software, or service. And with the growing size and scope of code, together with expanded libraries, APIs, and middleware, the attack surface of modern code is vast.

Anthropic has let some developers test an early version of Mythos.

It's the job of security engineers at companies like Apple to find and fix all the potential security flaws, but it's too big a job. Hackers only have to find a single unknown flaw, while the security engineers have to find and fix all of them.

This gives the attackers a major advantage. It has meant that software security has become less about an attempt to close every hole than it is about raising the bar for attackers—making exploits so difficult and expensive that they at least become rare.

But all that is about to change.

AI CODING AGENTS ARE CHANGING THE RULES

AI coding agents have gotten really good. In fact, in many areas, they're better than your average programmer, and in some areas they're better than all but the best experts. Anthropic's AI model Opus and the Claude Code tool are considered among the best. The folks at Mozilla used Opus 4.6 to scan through the Firefox code base (fave.co/4uH09ue) and found 22 security-sensitive bugs.