Zoom Patches Device Hijacking Flaw Found With Fewer Than 20 AI Prompts

Zoom patched a screen-sharing vulnerability that researchers say allowed an attacker to take control of participants’ devices during a meeting. A Security disclosed the flaw on August 11, describing a path through Zoom’s annotation feature that could cross the boundary between a shared screen and protected applications. The Verge and Wired AI both reported that researchers found the issue with fewer than 20 prompts submitted to publicly available AI models.

The available reporting describes the disclosure as a warning about AI-assisted vulnerability research rather than evidence of a confirmed mass exploitation campaign. The Hacker News and The Next Web headlines also characterize the issue as a screen-sharing or screen-access flaw, while Computerworld reports that Zoom issued a patch for an account-takeover hole. The sources provide limited technical detail, so the precise attack sequence and affected versions remain dependent on Zoom’s security bulletin and A Security’s original disclosure.

The vulnerability centered on Zoom’s annotation tools, according to The Verge, which reported that an attacker could use the feature during a screen-sharing session to interact with or reach applications that should have remained restricted. Wired AI described the consequence more broadly: anyone on a call involving screen sharing could have had their device taken over. The Hacker News headline similarly says the bug allowed access to restricted applications.

Those descriptions point to a security boundary failure inside a familiar collaboration workflow. Screen sharing is designed to expose selected visual content, while annotations let other participants draw or interact with that presentation. When those controls are connected incorrectly to the operating system or protected applications, a meeting participant can gain far more access than the host intended. The Next Web framed the outcome as attackers hijacking screens during conference calls, and Computerworld’s account-takeover description indicates that Zoom treated the issue as a serious security defect.

The reported prompt count shows how quickly public AI systems can assist security research. A Security told The Verge and Wired AI that researchers needed fewer than 20 prompts to guide publicly available models toward the Zoom flaw. The result highlights the practical value of asking an AI system to inspect product behavior, identify trust boundaries, and propose attack paths, rather than treating AI as a fully autonomous hacker.

The finding also has limits. The sources don't identify the exact models, prompts, researchers’ tooling, or how much conventional testing followed the AI output. Wired AI places the Zoom disclosure alongside broader examples of models finding vulnerabilities and developing exploits, but the article's central fact remains narrower: AI helped uncover a flaw that Zoom then fixed. Medium provides no usable reporting in the supplied material, and Batwheels.fandom is unrelated, so neither supports conclusions about the incident.

Zoom has patched the vulnerability, according to The Verge and Computerworld. The Verge linked the fix to Zoom security bulletin ZSB-26015, while Computerworld reported the issue as an account-takeover hole. The available excerpts don't state whether exploitation occurred in the wild, how many users were exposed, or which operating systems and Zoom versions were affected.

Users and organizations should apply Zoom updates through their normal software-management process and review meeting controls that govern screen sharing and annotation. Administrators can also treat unexpected annotation activity as a security signal, especially in meetings involving sensitive systems. The disclosure matters because the vulnerable action could occur inside an ordinary call, where participants often trust built-in collaboration features. The Dark Reading source concerns BlueNoroff using fake Zoom calls as attack lures, adding a separate reminder that attackers already use the platform’s meeting context for social engineering, but it does not establish a connection to the patched bug.

The next step is independent validation of the patch and fuller technical disclosure. A Security’s report, Zoom’s bulletin, and follow-up coverage from Wired AI, The Verge, and Computerworld should clarify affected releases, exploitation requirements, mitigations, and whether customers needed action beyond updating the application. Those details will determine whether the incident was a contained product flaw or a broader lesson about desktop collaboration security.

For software makers, the episode puts AI-assisted testing into a concrete frame: a small number of prompts helped expose a defect in a widely used feature, while human researchers still had to reproduce, document, and disclose it. For defenders, the practical priority is equally concrete: install Zoom’s fix, keep meeting software current, and review what screen-sharing and annotation permissions expose. The claim is substantial, but the supplied sources don't support calling it a confirmed widespread attack.