A public artificial intelligence system required fewer than 20 prompts to uncover a severe vulnerability in Zoom's screen-sharing feature, researchers reported. The flaw allowed any participant on a call to hijack another user's device. Zoom has since deployed a fix.

What You Need to Know

The bug exploited Zoom's screen-sharing mechanism to enable unauthorized remote control of a target machine. It was identified using a widely available AI tool, not custom software. The patch closed the hole without requiring user action. The incident highlights how artificial intelligence is accelerating the discovery of critical vulnerabilities.

The Discovery Process

Call Researchers, a security analysis group, uncovered the flaw. They labeled the vulnerability 'A Zoom Screen' and described it as 'Sharing Bug Let Anyone Take Over Other Devices'. The team used a publicly accessible AI model to probe Zoom's codebase. Within fewer than 20 iterative prompts, the tool identified the exploit path. The researchers then reported the issue to Zoom under a coordinated disclosure timeline.

Technical Details

The vulnerability resided in Zoom's screen-sharing stream handling. An attacker on the same call could inject malformed data into the sharing session, which the receiving client processed with elevated privileges. This allowed the hijacker to execute arbitrary commands on the victim's device. The attack required no additional software or user interaction beyond joining the call. Potential exploitation scenarios included:

  • Remote command execution: Attackers could run malicious code on a hijacked system.
  • Data exfiltration: Sensitive files and credentials could be stolen silently.
  • Persistence installation: The takeover enabled backdoor deployment for ongoing access.

AI and the Changing Attack Surface

The use of a generic AI tool to find a zero-day vulnerability in under 20 attempts marks a shift in the security landscape. Traditional bug hunting often requires deep domain expertise and hours of manual code analysis. AI models can now automate pattern recognition and exploit synthesis, lowering the barrier for discovering serious flaws. This development pressures software vendors to adopt automated code auditing and faster patch cycles. It also raises questions about the potential for malicious actors to replicate the same techniques at scale.

Why This Matters

For enterprises and remote workers who depend on Zoom daily, this vulnerability demonstrates that even mature platforms can harbor critical weaknesses. The rapid discovery via a public AI tool means that similar flaws could be found and exploited before vendors respond. Organizations must reassess their tolerance for unknown risk and invest in proactive security measures such as endpoint detection and regular penetration testing. As AI-driven vulnerability research grows more accessible, the window between disclosure and exploitation will shrink, making timely patching a competitive necessity.

Zoom has not disclosed whether any active exploitation occurred before the fix. The company continues to recommend users keep their software updated. The incident serves as a reminder that in an era of AI-powered discovery, security is a moving target.