AI Researchers Uncover Zoom Bug That Could Let Call Participants Seize Devices

Security researchers have disclosed a Zoom vulnerability that, in the worst case, could have let an attacker take over a person’s device during a screen-sharing call with no warning and no victim interaction. Zoom says it has already begun rolling out fixes for the flaw, which affected supported versions of Windows, macOS, Linux, iOS, and Android.

The finding is significant because it shows how publicly available AI tools are accelerating bug discovery: researchers at the digital defense firm A Security say they uncovered the issue in early June using fewer than 20 prompts. The case is an early example of how AI-assisted security research is shrinking the time and skill needed to find serious weaknesses in widely used software.

What happened in Zoom?

The issue centered on the real-time annotation feature used during screen sharing. According to the researchers, the weakness could have allowed an attacker on the same call to exploit the protocol and gain control of a target device without any clear sign to the user.

That makes the flaw especially concerning. Joining a Zoom meeting is usually treated as a routine, trusted action, whether the call is private, internal, public, or part of a webinar. The researchers argued that this built-in trust is exactly what would make the vulnerability dangerous in practice.

A Security cofounder Omer Gull said the broader risk is that the tools for finding bugs are becoming far more accessible, lowering the barrier for would-be attackers and defenders alike. He added that a target like Zoom is especially valuable because users generally do not perceive the platform as threatening.

Zoom did not publicly respond to multiple requests for comment from WIRED about the findings, but it did publish a security advisory on Tuesday and said it has already started deploying fixes.

How did AI help uncover the flaw?

AI made the discovery faster, more automated, and more accessible. The researchers said their systems identified the weakness after fewer than 20 prompts to publicly available AI models, then used the same tools to build a working attack path.

In traditional security research, finding a flaw of this kind could require a small team working for months, carefully reverse-engineering software behavior, testing edge cases, and refining exploit logic. A Security’s account suggests that AI tools can now compress that cycle dramatically.

Why does that matter for defenders?

It matters because the same capability that helps defenders can also help attackers. The researchers warned that the cost of discovering and weaponizing software bugs is falling fast, and that shift could expand the number of people capable of carrying out advanced attacks.

Security experts often describe vulnerability research as a cat-and-mouse contest. The new reality, they say, is turning that contest into a race: defenders want to patch faster than attackers can discover and exploit flaws, while AI tools are speeding up both sides.

Which systems were affected?

The vulnerability touched Zoom’s supported operating systems across desktop and mobile platforms. Zoom said the issue affected Windows, macOS, Linux, iOS, and Android, making it a cross-platform problem rather than a flaw isolated to one environment.

That broad reach raises the stakes. When a bug spans multiple operating systems, it can affect both enterprise deployments and consumer use, including remote work, sales calls, customer support sessions, classes, and public events.

Why screen sharing created such a risky target

Screen sharing is a trusted activity by design. A user joins a meeting, sees coworkers or clients, and often gives little thought to the underlying software pathways that make collaboration possible. That normal sense of trust is what made the researchers worry about the bug’s real-world consequences.

Zoom is also widely used for webinars, town halls, public briefings, support sessions, and other semi-public settings. In those environments, a malicious participant could potentially blend in more easily than on a highly restricted corporate call.

What makes obscure features dangerous?

Obscure or specialized features are often harder to scrutinize than core functions. The researchers said they targeted the annotation system because complex, less obvious components can contain overlooked mistakes, especially in proprietary software that is not open to outside review.

That logic mirrors the approach of human bug hunters, who often focus on less-traveled code paths. The difference now is that AI can explore those paths much faster, which may make overlooked software corners less safe than they once were.

How could an attacker have used it?

An attacker who successfully exploited the flaw could have potentially taken control of a victim’s device during the call, then used access to steal credentials or move deeper into an organization’s network.

That is the nightmare scenario for enterprise security. One compromised laptop can become a bridge into email, document systems, authentication tools, internal chat platforms, and cloud infrastructure.

Yossi Torati, also a cofounder of A Security, told WIRED that if an attacker had the vulnerability in hand, they could potentially join a call with a company employee, seize control of that computer, obtain credentials, and then pivot through the organization.

The researchers said the issue was silent and required no meaningful user interaction. That is the kind of vulnerability security teams fear most because it bypasses many of the warning signs people rely on, such as suspicious links, unusual attachments, or software prompts.

What has Zoom done so far?

Zoom says it has already begun remediation and released a security advisory describing the issue. The company has deployed both server-side and client-side fixes, meaning the patch work covers Zoom’s own infrastructure as well as the apps installed on user devices.

That dual approach is important because modern software vulnerabilities often require changes both in the cloud and on endpoints. If one side is fixed and the other is not, attackers can sometimes find a path through the gap.

Zoom has not publicly explained the technical details of the exploit beyond the advisory, but the company’s response suggests the issue was considered serious enough to warrant immediate patching across all major platforms.

Why is this part of a larger AI security shift?

This disclosure is more than a single product bug report. It is evidence that AI is changing how software flaws are discovered, tested, and weaponized.

For years, security researchers have used automation, fuzzing, static analysis, and reverse engineering tools to find vulnerabilities. AI models now add another layer: they can reason about code and behavior, suggest promising attack directions, and iterate quickly through thousands of hypotheses.

That doesn’t mean AI automatically finds every bug. But it does mean that some categories of vulnerabilities, especially in complicated or poorly documented code, may now be easier to surface than before.

What is the democratization risk?

The democratization risk is that advanced vulnerability discovery is no longer confined to highly specialized teams. If a small group can do in hours what used to take months, then more people—including malicious actors—may gain access to sophisticated techniques.

That shift could put pressure on vendors to patch faster, on companies to tighten meeting security, and on users to treat even familiar collaboration apps with more caution.

How should organizations respond?

Organizations should assume that mainstream collaboration tools can be targets, not just endpoints or browsers. Security teams may want to review patch status, enforce update policies, and keep screen-sharing privileges limited to what employees actually need.

A few practical steps can reduce exposure:

• Keep Zoom and all collaboration apps updated promptly.
• Restrict screen-sharing and annotation permissions to trusted users.
• Monitor meetings that involve external participants.
• Use least-privilege account controls and strong authentication.
• Train employees to report unusual behavior during calls immediately.

For many companies, the biggest lesson is not that Zoom is uniquely unsafe. It is that any widely trusted communications platform can become a high-value target when attackers can use AI to probe it faster than before.

Timeline of the disclosure

The sequence of events shows how quickly AI-assisted research can move from discovery to public warning.

What does this mean for the future of hacking?

It means the economics of vulnerability research are changing fast. If AI can uncover a serious bug in a popular enterprise product with a handful of prompts, then the pace of discovery may outstrip the pace of human review in some areas.

That could lead to more frequent emergency patches, more pressure on software makers to secure complex features, and more concern about how quickly attackers can adapt AI tools for exploitation.

At the same time, the same trend may strengthen defenders. Security teams equipped with the right models and workflows could identify weaknesses earlier, harden code faster, and reduce the window of exposure. The race is on, and the stakes are no longer limited to niche systems or obscure software.

For Zoom users, the immediate message is simple: update the app, assume patches matter, and treat even routine meetings as potentially sensitive. For the security industry, the message is broader and more unsettling: AI is now part of the vulnerability landscape, and it is changing the speed of the game.