Calif, a security company based in Palo Alto, California, disclosed today that it built a self-spreading worm capable of hijacking WeChat accounts by exploiting a memory corruption flaw in the app's voice-over-IP stack — and did so with significant assistance from AI.
The worm, named WeWorm, requires no interaction from a target. Placing a call to a WeChat user is enough to compromise their account, whether or not they answer. Only declining the call within seconds of it ringing would prevent the exploit from succeeding, the company said. After seizing control of an account, the worm dials numbers from the victim's list to extend the attack outward.
According to the company, its researchers worked alongside AI to identify the vulnerability and produce an initial remote code execution exploit in approximately two days. The complete worm took one additional week to build, the company said. Calif chief executive Thai Duong told the New York Times that a project of this scale would previously have required a larger team and months of work.
Part of what makes the flaw so consequential is that WeChat extends elevated trust to phone numbers an account has saved, a design choice the worm turns into a propagation mechanism. Once one account is compromised, that trust relationship enables the worm to spread further. Combined with separate Android and iOS vulnerabilities that Calif said it has reported and is helping fix, the attack could lead to full control of a victim's device.
Calif reported the bug to Tencent, the Chinese technology company that owns WeChat, in July. Tencent issued updated versions of the app for both iOS and Android in August and later confirmed that the exploit had been mitigated on the server side for all users, meaning no user action is required. Tencent told the Times it had no reason to believe any users were affected.
WeChat has more than 1.4 billion monthly active users, the company said in a blog post this year, with the overwhelming majority based in China.
Calif said it is withholding technical details of the vulnerability and plans to present a full analysis at an upcoming conference. The company said it relied on a combination of open-source AI models and leading commercial models to conduct the research, but declined to specify which ones, according to the Times.
The security industry has seen a rapid rise in AI-assisted vulnerability research in recent months. Google $GOOGL 's Threat Intelligence Group identified a cybercrime group that used an AI model to discover and weaponize a previously unknown zero-day vulnerability — the first confirmed instance of that kind. Separately, OpenAI disclosed that its AI agents escaped a controlled testing environment and compromised parts of Hugging Face's production infrastructure , chaining together previously unknown vulnerabilities in the process.
Calif framed the disclosure as a public warning, arguing that AI has fundamentally shifted who can mount sophisticated attacks — tools once confined to richly resourced, advanced threat groups are now accessible to far less capable operators — and urged governments and the private sector to work together to turn AI toward defensive ends.
The essential business news, delivered fresh every morning.
Join 500,000+ readers who start their day with Quartz.
The full story
This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.
