WeWorm: AI built a zero-click WeChat exploit in 2 days

Security firm Calif built a zero-click worm for WeChat — an app with over a billion users — and the headline is not the worm. It is the clock. The team found a memory-corruption bug in WeChat's VoIP stack and had working remote code execution in roughly two days, with the self-propagating worm about a week later. AI did the heavy lifting on both the bug hunt and the initial exploit. That is the number every operator should be writing down.

What actually happened

The flaw sits in WeChat's voice-call handling. An incoming call from an attacker triggers memory corruption before the victim does anything — no tap, no answer. Help Net Security reports that the account is taken over while the phone is still ringing, and the code then reads the victim's contacts and calls them. Compromise to re-propagation, in seconds, with no human in the loop.

Calif reported it privately to Tencent, which shipped patched iOS and Android builds plus a server-side mitigation that protects users with no action on their part. Researchers withheld technical details and found no evidence of real-world abuse before the fix. Experts quoted in the coverage estimated the worm could have reached hundreds of millions of devices within hours if it had been released.

We are not naming a CVE here, because none was published in the reporting we could verify.

Why AI-accelerated exploit development matters for your business

You do not run WeChat's VoIP stack. You do run something with a parser in it — an image handler, a webhook receiver, a PDF pipeline, a media upload path. The transferable fact is that a small team compressed "find a memory bug in a hardened billion-user app" and "write working RCE" into about two days by pointing models at it.

Every patch-window assumption you have is priced off the old cost of exploit development. That price just fell. Two consequences:

• Time-to-exploit is collapsing, so time-to-patch has to. We wrote about the Adobe Commerce store that fell 50 minutes after the exploitation report. That is the shape of the new window. A monthly patch cadence is now a policy that describes last decade.
• Zero-click means user training does nothing. No amount of "don't click suspicious links" defends an account that is taken over mid-ring. Controls have to be structural: keep untrusted input away from native parsers, sandbox the ones you cannot avoid, and be able to revoke sessions and tokens fast when something lands.

The optimistic read is real, too. The same acceleration is available to defenders — the researchers here were the good guys, and Tencent's users were protected before anyone knew. But attackers get the tool on the same day you do, and they do not have a change advisory board.

Sources: Help Net Security, Security Affairs.