On Tuesday the California security firm Calif published its work on WeWorm. It’s a worm that spreads through WeChat calls, on iOS and Android, without the victim answering. Seconds later the account belongs to the attacker, and the phone starts calling the victim’s contacts. WeChat has 1.4 billion monthly active users.
Answering doesn’t help, by the way. Pick up and you hear nothing, and the exploit lands anyway. The bug itself is a memory corruption issue in WeChat’s VoIP stack. Calif is holding back the technical details for a conference talk.
The sentence that matters
“Working with AI, our team found the bug and wrote the first remote code execution (RCE) exploit in about two days.” The full worm took another week on top of that.
Calif frames it themselves: something at this scale used to take a larger team months. And: “AI can already do most of the work here. Our team provided the judgment about what to target and how to test it safely.”
Which model they used isn’t stated anywhere. A shame, but it doesn’t change the finding. Hunting memory bugs and getting a first working exploit is grinding, highly structured work — exactly the kind of task current models are good at.
The process was clean
- July 24: reported to Tencent
- August 21: patched builds ship (Android 8.0.77, iOS 8.0.76)
- August 28: server-side mitigation confirmed for all users
- September 8: public disclosure
Tencent validated the findings. If you use WeChat and have updated since August, you’re fine.
Why this is on a Claude blog
Every debate about AI and security eventually reaches the argument that defenders benefit as much as attackers. That’s fair — Calif are the good guys here. They reported, they waited, they published after the patch.
The distribution isn’t symmetrical, though. An attacker needs one hole. A defender has to find all of them. When the effort for a first exploit drops from months to days, that math moves, and not in defense’s favor. Calif say as much themselves: the bar drops for less experienced actors too.
That’s why I find this more interesting than most benchmark results. Someone wrote down how long a thing took that used to take much longer. It’s a harder number than any score.
Sources: Calif: WeWorm · The Register: WeChat worm could pwn a friend before they even answered the call · Help Net Security: ‘Zero-click’ WeChat worm could hijack accounts · Simon Willison: Calif Research