Israeli cybersecurity firm Ⓐ Security disclosed “Zoomsday”, a critical zero-click vulnerability in Zoom that allowed any meeting participant to take complete control of every other attendee’s device through a single malicious annotation message.
The flaw affected every Zoom Workplace client across Windows, macOS, iOS, Android, and Linux in all versions up to and including 7.0.5, and the entire discovery from initial research through working exploit took fewer than 20 AI prompts and under 24 hours.
The vulnerability exploited Zoom’s annotation feature, the tool that lets participants draw, type text, add shapes, or place stamps on a shared screen during a meeting. Annotation runs on a proprietary protocol with no public documentation, and every Zoom client automatically parses whatever it receives through this channel without validating the sender’s role or message type.
Ⓐ Security’s researcher Idan Levcovich discovered that the protocol’s factory dispatcher could not distinguish between a legitimate acknowledgement message and a fully armed object payload, meaning an attacker could embed malicious code inside what the receiver’s parser treated as a routine response.
The core bug was a memory-corruption flaw tracked as CVE-2026-53413 with a critical CVSS 4.0 score of 9.0. The annotation protocol’s text deserializer reads four fixed 128-byte buffers using a 32-bit character count supplied directly from the wire, with no bounds check comparing the count against the buffer’s actual capacity. An oversized count writes past the buffer’s boundary into the function’s stack frame, overwriting saved registers and the return address with attacker-controlled bytes. A companion over-read vulnerability (CVE-2026-53414) leaked live heap contents including code pointers and vtable addresses, providing the ASLR bypass material an attacker needs to target the overflow precisely.
A third independent vulnerability (CVE-2026-53415) involved a separate message type that read attacker-supplied bytes directly into an internal linked-list structure and then unlinked it without validating the pointers, creating a write-what-where primitive offering another path to remote code execution.
The attack required nothing from the victim beyond being present in the meeting, making it a true zero-click exploit with no visual indicator of compromise. From a presenter seat, the attacker held a direct channel to every participant individually. Which meant one malicious annotation could cascade across an entire room of targets. Once code execution was achieved, the attacker could quietly steal personal data, activate the microphone or camera for surveillance, install additional malware, or pivot deeper into the victim’s network.
The AI dimension is what elevates this disclosure from a severe bug report into an industry-level warning about the collapsing cost of offensive capabilities.
Ⓐ Security’s AI agent mapped the attack surface, reverse-engineered the closed-source annotation protocol, identified the memory corruption vulnerability, built the exploit payload, and achieved confirmed remote code execution on a live victim, all using publicly available frontier models in a single working day. This class of exploit discovery previously demanded nation-state infrastructure, elite research teams, months of dedicated effort, and budgets that governments regulate as weapons exports. That barrier no longer exists.
Zoom responded quickly after receiving the report on June 10, deploying a client-side fix in version 7.1.0 by June 22 and a server-side mitigation on July 15 that filters malicious annotation messages before they reach clients.
However, the server-side protection does not cover end-to-end encrypted meetings because the server cannot inspect content it cannot decrypt. Organizations using Zoom E2EE with clients older than 7.1.5 remain exposed and should update immediately.
