Zoom faced a serious security issue this week after researchers exposed a chain of flaws in its annotation tools. The problems allowed zero-click remote code execution during meetings. Attackers could join or host sessions and target other participants without any action from victims. No clicks, downloads, or visible signs appeared on affected devices.
A Security identified the chain and named it Zoomsday. Their team built a working exploit in under 24 hours. They used fewer than 20 prompts on public AI models. The firm described the lowered barrier in stark terms. “The barrier to producing this class of weapon has collapsed, and it won’t come back,” researchers stated in their disclosure.
The vulnerabilities centered on how Zoom clients processed data through a proprietary annotation protocol. Three CVEs formed the core: CVE-2026-53413 involved memory corruption that supported code execution. CVE-2026-53414 allowed memory over-reads that leaked addresses and helped bypass protections like ASLR. CVE-2026-53415 created a write-what-where condition in auto-shape metadata handling. Successful attacks could run malicious code, access cameras or microphones, steal data, or install further malware. The chain hit Windows, macOS, iOS, and Android clients.
Zoom released patches for the issues. A server-side filter blocked malicious messages in standard meetings. End-to-end encrypted sessions lacked that protection because servers could not inspect content. Organizations received advice to update clients promptly. Additional steps included restricting annotation features, using waiting rooms, and monitoring endpoints for unusual activity.
Reports from SecurityWeek detailed the zero-click nature and confirmed the annotation vector as the entry point. The Verge noted the rapid AI-assisted discovery process. eSecurityPlanet highlighted the same timeline and prompt count. SecurityAffairs covered the patches and platform-wide reach. These accounts align on the exploit’s mechanics and the role of public AI tools in shortening development time.
The episode shows how AI compresses the window between vulnerability discovery and functional weaponization. Memory corruption issues in collaboration software have long existed. Automatic parsing of incoming data creates persistent attack surfaces. Here the combination proved especially potent because any meeting participant could serve as attacker or target. Presenters and viewers faced equal exposure through direct client-to-client channels.
Zoom urged users to apply the updates. Broader guidance emphasized inventories of installed software and least-privilege policies on endpoints. Incident response testing with similar scenarios became a recommended practice. The fixes closed the immediate paths, yet the research underscores ongoing pressure on vendors to harden annotation and sharing features across platforms.
Real-world impact remains contained so far because patches arrived quickly. Still, the findings serve as a reminder that widely deployed meeting software carries high stakes when flaws surface. AI assistance in research accelerates both defensive and offensive work. Defenders must match that pace with faster patching cycles and layered controls.