Zoom Patches AI-Discovered Flaw That Could Let Attackers Hijack Calls

Zoom has fixed a serious vulnerability that security researchers say could have let an attacker take control of devices belonging to people on a meeting call, without the victims doing anything to trigger it. The flaw, found with the help of fewer than 20 prompts to public AI models, affected Zoom across Windows, macOS, Linux, Android and iOS, underscoring how generative AI is now accelerating both offensive security research and the hunt for software weaknesses.

The issue centered on Zoom’s annotation feature, which lets participants draw or write on shared screens. According to researchers, an attacker could use the weakness to execute malicious code on other meeting participants’ devices, potentially enabling data theft, camera or microphone access, or malware installation. Zoom says it has already shipped a fix.

What happened, and why does it matter?

The newly disclosed bug is notable not just because it affected a widely used workplace app, but because of the speed with which it was identified. Researchers at A Security say they were able to uncover the exploit using AI-assisted prompts rather than the kind of long, resource-heavy effort traditionally associated with finding high-end software vulnerabilities.

That matters because Zoom is deeply embedded in business, education and government workflows. A compromise in a meeting platform can expose sensitive discussions, internal documents and device-level data in one stroke. In practical terms, a flaw that gives an attacker code execution on a participant’s machine is far more serious than a typical meeting disruption or chat prank.

The discovery also highlights a broader shift in cybersecurity. Tools powered by public large language models are lowering the barrier for advanced vulnerability research, making techniques once reserved for elite teams more accessible to smaller groups — and potentially to criminals.

How did the Zoom flaw work?

The flaw involved Zoom’s annotation tool, a feature designed for collaboration during screen sharing. In a normal meeting, annotation lets users mark up a shared screen with notes, arrows or drawings. In the vulnerable version, researchers found a path that could be abused to run code on a target device.

Once exploited, the attacker could join or host a meeting and pivot from the meeting environment into the victim’s system. That could open the door to stealing files, spying through the webcam or microphone, or dropping malware for later use. Crucially, the researchers said the attack did not require the target to click a link, accept a file, or otherwise take a visible action.

Perhaps most concerning, the compromise left no obvious warning on the victim’s screen. In other words, a person could be participating in what appears to be a routine call while their device is being silently manipulated in the background.

Why is “no visual cue” such a big deal?

It is a big deal because security defenses often rely on users spotting something suspicious. If an exploit produces no visual clue, no pop-up, and no obvious crash, the attacker can remain hidden longer. That increases the risk of data exfiltration, unauthorized surveillance and persistence on the machine.

For enterprise users, stealthy compromise is especially dangerous because it can spread from one meeting attendee to a broader corporate network if credentials or session data are harvested.

How AI helped find the vulnerability

A Security says it identified the weakness using fewer than 20 prompts on publicly available AI models. The firm’s vulnerability researcher, Idan Levcovich, described the work as the kind of thing that once required nation-state-scale resources, specialized personnel and months of effort.

Instead, the researchers say they used an AI agent to accelerate the process and reach a working exploit in a single day. That is an important detail: the tool was not magically discovering a flaw on its own, but it materially shortened the path from hypothesis to proof of concept.

“Producing a working exploit against it has always been nation-state work: elite teams, months of effort, budgets that governments regulate as weapons,” one researcher at A Security wrote. “A Security did it in a single day, with an AI agent and models anyone can access today.”

This is the kind of statement that security professionals will likely debate for months. Some will argue that AI merely speeds up an already well-established workflow. Others will see it as evidence that the defensive and offensive sides of vulnerability research are being reshaped in real time.

Is AI actually making hackers more dangerous?

AI is not replacing technical skill, but it is amplifying it. Public models can help with code analysis, pattern recognition, exploit brainstorming and iterative testing. That means a smaller team can do work that previously demanded more manpower and time.

In practice, that can help defenders find bugs faster too. But it also lowers the cost of entry for attackers, especially “script kiddie” style operators who can stitch together guidance, sample code and automation into something more harmful than they could produce alone.

Which devices and systems were affected?

The patched vulnerability affected Zoom on Windows, macOS, Linux, Android and iOS, according to the company’s Tuesday fix. That broad footprint is important because it means the issue was not limited to one desktop operating system or a niche mobile build.

When a flaw spans multiple platforms, enterprise security teams cannot rely on a partial rollback or a single operating system workaround. They need to verify patch status across the entire fleet, including remote workers, shared devices and personal phones used for business meetings.

Why Zoom’s annotation feature became an attack surface

Features built for convenience often become security liabilities when they interact with complex input, rendering or collaboration systems. Annotation tools are especially sensitive because they sit at the intersection of user interaction, remote content sharing and code paths that must interpret what participants are drawing.

In a conferencing application, a small mistake in how those inputs are processed can have outsized consequences. Attackers often look for precisely these kinds of features because they are widely deployed, frequently updated and difficult to audit across every edge case.

Zoom is hardly alone. Collaboration platforms tend to accumulate powerful features over time — breakout rooms, file sharing, remote control, virtual backgrounds, whiteboards and annotation tools among them. Each new layer expands the attack surface.

What does this say about modern app security?

It shows that even mature products can contain bugs that are both technically sophisticated and operationally dangerous. The issue is not only whether a company ships security updates, but whether product design has enough guardrails to keep a niche feature from turning into a system compromise.

It also reinforces a familiar cybersecurity lesson: if a feature allows remote interaction with code, graphics or input streams, it deserves the same scrutiny as any other exposed interface.

How should users and IT teams respond?

The immediate answer is straightforward: update Zoom everywhere it is installed. Because the flaw affected multiple operating systems, organizations should assume the patch needs to be rolled out broadly, not just to office desktops.

IT teams should also verify whether managed mobile devices received the fix, and whether employees use personal phones or tablets for meetings. In many companies, those devices are the weak link because they fall outside standard patch management.

Beyond patching, security teams should review whether meeting platforms are isolated from critical systems, especially on laptops used to handle sensitive work. A conferencing app should not have an easy path to privileged access or broad file exposure.

• Confirm the latest Zoom version is installed on every supported platform.
• Check whether endpoint protection flagged any unusual activity around meetings.
• Review permissions for microphones, cameras and screen-sharing tools.
• Remind staff not to install unofficial add-ons or meeting utilities.
• Use network segmentation and least-privilege access where possible.

Why this disclosure is a milestone for AI and cybersecurity

This incident is likely to be cited as an example of how quickly AI is changing vulnerability research. The notable part is not that researchers used automation — they have for years — but that the work reportedly moved from idea to exploit with unusual speed using tools available to almost anyone.

That matters because the security field has long assumed a rough asymmetry: attackers have more time, defenders have more complexity, and the most damaging bugs are expensive to find. If AI compresses that timeline, then the gap between discovery and abuse can shrink dramatically.

For software vendors, that means the standard patch-and-disclose cycle may become less forgiving. Vulnerabilities that once might have remained hidden for months could now be exposed faster, exploited sooner and weaponized more broadly.

For users, it is another reminder that the software supporting everyday work can fail in ways that are invisible until after the damage is done.

What security researchers are watching next

The most immediate question is whether other collaboration platforms have similar weaknesses in their shared-content or annotation systems. Researchers will likely examine whether AI-assisted workflows can uncover classes of bugs faster in products with comparable feature sets.

Security teams are also likely to look more closely at the use of public AI models in vulnerability research. If a simple prompt sequence can help expose a major flaw, then the same methods may be used by both ethical researchers and malicious actors — a dual-use reality that is becoming impossible to ignore.

In the short term, Zoom’s patch closes one dangerous path. In the longer term, the disclosure points to a broader and more unsettling trend: the tools that help developers and researchers move faster are now helping attackers move faster too.

Timeline of the disclosure

The bigger picture

Zoom’s vulnerability is a product security story, but it is also an AI story. It shows how accessible models can help uncover dangerous bugs much faster than the industry is accustomed to, and how that speed can cut both ways. Defensive research benefits, but so do offensive operators.

For now, the practical takeaway is simple: Zoom users should make sure their apps are updated. The strategic takeaway is more unsettling. As AI becomes a standard tool in cybersecurity, the line between advanced research and mass-access exploitation will likely get thinner.

That is why this disclosure matters far beyond one meeting app. It is a glimpse of what happens when consumer-facing AI tools collide with enterprise security at scale.

And it is a warning that the next serious flaw may not stay hidden for long.

Frequently asked questions

What was the Zoom vulnerability?

It was a security flaw in Zoom’s annotation feature that researchers say could allow an attacker to run malicious code on a meeting participant’s device. The issue could enable data theft, surveillance through the camera or microphone, and malware installation.

Did victims need to click anything for the attack to work?

No. Researchers said the attack required no action from the victim and did not show an obvious visual warning. That makes the exploit especially concerning because users could be compromised during a normal-looking meeting.

Which platforms were affected?

The vulnerability affected Zoom on Windows, macOS, Linux, Android and iOS. Zoom has since issued a fix across those platforms, so users should update their apps as soon as possible.

How did researchers find the bug so quickly?

Researchers at A Security said they used fewer than 20 prompts on publicly available AI models and reached a working exploit in a single day. Their report suggests AI tools significantly sped up the research process.

Why is this important beyond Zoom?

It shows that AI-assisted research can shorten the time needed to discover and weaponize vulnerabilities. That has implications for every software company, because flaws may be found and exploited much faster than before.