Zero-Click AI Worm: "WeWorm" Hijacks WeChat Accounts and Propagates via Audio Calls
Publication Date: September 9, 2026
Source Analysis: Calif.io Research | Help Net Security | Quartz
|---|Author: DigitalSpying Research Team
Executive Summary
Cybersecurity researchers at Palo Alto-based security firm Calif have publicly disclosed "WeWorm," the first documented proof-of-concept zero-click, self-propagating worm designed to compromise accounts and spread across the global WeChat messaging network. Unlike conventional messaging malware that requires user interaction—such as clicking on malicious phishing links or downloading suspicious attachments—WeWorm spreads autonomously over incoming voice or video call streams without requiring the recipient to answer or touch their device.
The worm exploits a critical heap memory corruption vulnerability in WeChat’s underlying WebRTC/VoIP media processing engine across both Apple iOS and Google Android platforms. In demonstration testing, an autonomous offensive AI agent generated the precise multi-stage exploit chain. Once a target device receives an initial audio call signal, the payload executes, extracts session authentication tokens, accesses camera and microphone hardware, and silently commands the WeChat application to initiate outbound exploit calls to all recent contacts, creating an exponential infection cascade.
WeChat’s parent company, Tencent, has deployed emergency server-side filtering rules and rolled out hotfix client updates following responsible disclosure.
Deep-Dive Technical Analysis
1. The Zero-Click Ingestion Vector
WeChat’s VoIP calling subsystem relies on custom real-time transport protocol (RTP) and Session Description Protocol (SDP) packet parsers to establish peer-to-peer audio and video streaming sessions.
* Packet Processing Before Answer: When a caller dials another WeChat user, the receiving client must process incoming SDP offer packets and pre-negotiate audio codecs (such as Opus or AAC) before the recipient decides to accept or decline the call. This occurs automatically in the background as the phone rings.
* Heap Buffer Overflow in Codec Parser: WeWorm targets a classic integer truncation leading to a heap buffer overflow in WeChat’s custom audio frame depacketizer. By embedding a stream of oversized, fragmented RTP extension headers into the initial ringing packet sequence, the media library allocates insufficient heap memory and proceeds to write incoming packet payload data past the boundary of the allocated buffer.
2. The AI-Engineered Exploit Chain
Researchers at Calif revealed that the exploit harness was developed using an automated offensive AI reasoning model:
* Dynamic ASLR & Sandbox Defeat: The AI agent automated the heap-feng-shui arrangement required to align heap chunks in the constrained mobile sandbox. On iOS, it defeated Address Space Layout Randomization (ASLR) by querying predictable memory structures in shared dynamic caches; on Android, it leveraged a secondary information disclosure in the logging daemon to leak base memory addresses.
* In-Process API Hooking: Once arbitrary code execution is achieved, the payload does not terminate the app or disrupt the user interface (preventing crashes). Instead, it hooks WeChat’s internal objective-C / Java messaging runtime (MicroMsg framework) in memory.
* Autonomous Worm Propagation:
1. The worm queries the local encrypted SQLite database to dump the victim's session tokens, contact list, and recent call history.
2. It instructs the application’s VoIP service to place silent, background VoIP calls to the top 20 active contacts on the user’s contact list.
3. During each outbound call attempt, it sends the weaponized RTP packet sequence.
4. The remote recipients' devices are compromised immediately upon receiving the call signal, repeating the cycle recursively across the network.
Industry Impact & Mitigation Strategies
Strategic Implications
* Rise of AI-Synthesized Exploits: WeWorm represents a major proof-of-concept demonstrating how offensive AI agents can compress the time required to weaponize complex heap vulnerabilities into robust, cross-platform zero-click worms.
* Critical Exposure of Mobile Messaging Infrastructure: With over 1.3 billion active users, messaging ecosystems represent single points of systemic failure. Self-propagating zero-click vulnerabilities in consumer messaging apps pose unprecedented national security, espionage, and surveillance risks.
Defensive Countermeasures & Best Practices
1. Immediately Update the WeChat Client: Ensure the WeChat application is updated to version 8.0.52 (or later) on both iOS (via the Apple App Store) and Android (via Google Play or official vendor distribution).
2. Enable Device Lockdown / Advanced Protection Modes: Users operating in high-threat environments (journalists, executives, political figures) should enable Apple Lockdown Mode on iOS or Android Private Space restrictions, which disable non-essential media codecs and strict JIT compilation.
3. Restrict Unsolicited Voice/Video Calls: Configure in-app privacy settings to reject or mute calls from non-contacts:
* Navigate to: Me > Settings > Friends’ Permissions / Privacy and enforce strict contact-only connection policies.
4. Corporate Mobile Device Management (MDM): Enterprise security administrators should enforce mobile OS patching baselines and restrict consumer messaging apps on devices accessing enterprise email or corporate internal intranets.