Zoomsday is the name given to a critical Zoom flaw, publicly disclosed on August 11, 2026, that let any meeting participant take control of another attendee's device without a single click from the victim. What makes this case remarkable for an SME leader is not just the flaw itself: it was discovered and turned into a working exploit in under 24 hours, using fewer than 20 requests sent to publicly available AI models. Here are the sourced facts, and what to do right now.
At a glance
- Zoomsday (CVE-2026-53413) is a critical "zero-click" flaw in Zoom's annotation feature: a meeting participant could remotely execute code on another participant's device with no action required from them (source: Zoom, bulletin ZSB-26015).
- The flaw was found by research firm Ⓐ Security, which used publicly available AI models to reverse-engineer the annotation protocol and build a working exploit in under 24 hours, with fewer than 20 prompts (source: Ⓐ Security blog).
- Zoom fixed the issue on the client side on June 22, 2026 (12 days after the report), ahead of a coordinated public disclosure on August 11, 2026, alongside three related flaws (CVE-2026-53414, -53415, -53416).
- The fix is available from Zoom Workplace 7.1.5 (or 7.0.6 on the older branch), Zoom Rooms 7.1.5 and Meeting SDK 7.1.5: every organization should check that its machines are up to date.
- This episode comes alongside another alert the same week: a flaw in LiteLLM, a widely used tool for connecting applications to AI models, exposed more than 2,500 companies according to research firm CloudSEK. Two signals pointing the same way: AI infrastructure and everyday tools are now targets in their own right.
What is the Zoomsday flaw?
Zoomsday is a zero-click remote code execution (RCE) vulnerability located in Zoom's annotation feature, the tool that lets a participant draw, highlight, or type on a shared screen. The technical cause is a missing bounds check when processing annotation data: an attacker could send specially formatted data that overflows a memory buffer, executing their own code on other participants' devices.
In practice, an attacker only needed to be present in a Zoom meeting, on any operating system (Windows, macOS, Linux, iOS, Android), to take over another device: no click, no download, no action required from the victim, other than being in the meeting (source: Ⓐ Security).
Three related flaws were disclosed at the same time:
- CVE-2026-53414: excessive memory read (information disclosure), medium severity.
- CVE-2026-53415: use-after-free flaw, high severity.
- CVE-2026-53416: path traversal, limited to virtual desktop infrastructure (VDI) environments.
How AI made record-time discovery possible
What sets Zoomsday apart from a typical security flaw is how it was found. Ⓐ Security did not use a confidential, specialized tool: the team guided publicly available AI models through a series of steps, in natural language, to arrive at a working exploit.
Reverse-engineering the protocol
Mapping the attack surface
Building the payload
Validating the exploit
Key takeaway
This speed does not mean AI "invents" flaws out of nothing: it accelerates work that human experts already know how to do, over days or weeks. The real change is the timeline, which now approaches that of a motivated attacker rather than only a specialized research lab.
Timeline: from discovery to public disclosure
More than two months passed between discovery and the public announcement, the time it took Zoom to fix the issue before technical details were made public, a standard practice known as responsible disclosure.
June 8-9, 2026
Discovery and exploit
June 10-11, 2026
Report to Zoom
June 22, 2026
First client fix
July 15, 2026
Server-side mitigation
July 20, 2026
Full fix
August 11, 2026
Public disclosure
A broader signal: AI is also accelerating flaw research
Zoomsday is not an isolated case this week. Research firm CloudSEK revealed, the same day, a software supply-chain flaw affecting LiteLLM, a widely used open-source tool for connecting enterprise applications to AI models (OpenAI, Anthropic, and others). According to CloudSEK, the compromise of LiteLLM packages in early 2026 potentially exposed more than 2,500 organizations and 434,000 CI/CD pipelines, risking the theft of AI API keys, cloud credentials (AWS, Google Cloud, Azure), and access tokens.
These two announcements, published a day apart, point to an underlying trend: AI is now an accessible vulnerability-research tool, capable of speeding up the work of defenders (as here, through responsible disclosure) as much as that of less scrupulous attackers. And AI infrastructure itself (model-connection tools, open-source dependencies) is becoming a prime target, on par with video-conferencing or office software.
| Before (manual research) | With AI in 2026 |
|---|---|
| Discovering a complex critical flaw: several days to several weeks | Zoomsday: under 24 hours, fewer than 20 prompts |
| Skill required: specialized reverse engineering, low-level memory expertise | Natural-language guidance of a public AI model |
| High barrier to entry: reserved for specialized teams | Lower barrier: accessible to smaller teams, including malicious ones |
| Comfortable patching window for the vendor | Shorter patching window before an attacker reproduces the result |
What SMEs should do now
For an SME using Zoom daily, four concrete actions, in order of priority:
- Update immediately every machine to Zoom Workplace 7.1.5 (or 7.0.6), Zoom Rooms 7.1.5, and Meeting SDK 7.1.5.
- Set a mandatory minimum client version for every collaborator, including external guests, rather than relying on individual updates.
- Reduce the attack surface: enable waiting rooms and passcodes, disable annotation if it is not needed, prefer the browser client (sandboxed, without the annotation feature) for sensitive meetings.
- Extend vigilance beyond Zoom: if your company uses AI-connection tools like LiteLLM or similar gateways, check for available patches and consider rotating API keys as a precaution.
These reflexes echo the ones already recommended for securing AI agent connectors in the enterprise: systematic updates, least privilege, and access monitoring.
FAQ
Should I uninstall Zoom in the meantime?
No. The flaw has been fixed since versions 7.1.0 (initial fix) and 7.1.5 (full fix). The priority action is to check that every machine in the company has been updated, including those of remote workers or meeting-room equipment.
Why did Zoom wait two months to announce the flaw publicly?
This is standard practice in cybersecurity, known as responsible disclosure: the vendor fixes the issue before technical details are made public, so as not to hand attackers a blueprint while users are still vulnerable.
Did AI "invent" this flaw on its own?
No. Ⓐ Security's researchers guided the AI models at every step, with successive prompts. The AI accelerated reverse-engineering work that human experts already know how to do, sharply reducing the time and expertise required.
Should my company worry about the LiteLLM flaw even if it doesn't use that tool directly?
If your SME does not build applications connected to AI models through LiteLLM, the direct risk is low. That said, the incident illustrates a useful principle for any AI tool purchase: ask vendors which open-source dependencies they use and how they manage their security patches.
In conclusion
Zoomsday is a reminder of a simple but often overlooked rule for SMEs: software updates are not a matter of convenience, they are a front-line defense. The speed AI now brings to vulnerability research should not cause undue alarm, as long as basic reflexes stay current: patches applied quickly, a reduced attack surface, and vigilance over third-party tools connected to company systems. To go further on securing AI tools in the enterprise, browse our AI resources for business leaders.
