Zoomsday: AI Built a Zoom Zero-Click Exploit in Under 24 Hours — Patch Now

A Security researcher used fewer than 20 prompts with publicly available AI models to turn a critical Zoom vulnerability, CVE-2026-53413, into a working zero-click exploit in under 24 hours, dubbed 'Zoomsday,' which executes arbitrary code on any meeting participant's device. Zoom patched the flaw and three related CVEs on August 11, 2026, in security bulletins ZSB-26015 through ZSB-26018, and users are urged to update to Zoom Workplace 7.1.5 or 7.0.6, Zoom Rooms 7.1.5, Meeting SDK 7.1.5, and VDI Client 7.0.11 or 6.6.16. The researcher noted that AI eliminated months of manual exploitation work, though a human still had to point the AI at the vulnerable function, and OpenAI restricted its most advanced offensive-security model, GPT-5.6-Cyber, to vetted partners the day before disclosure.

A researcher at A Security turned a critical Zoom vulnerability into a working exploit in under 24 hours using fewer than 20 prompts with publicly available AI models. The result — dubbed “Zoomsday” — executes arbitrary code on any meeting participant’s device with zero clicks required from the victim. Zoom patched it August 11. If you haven’t updated, do that before reading further. What Zoomsday Actually Does The vulnerability chain centers on CVE-2026-53413 https://www.securityweek.com/zoom-patches-zero-click-code-execution-vulnerability/ , a critical flaw in Zoom’s screen-sharing annotation protocol. The bug lives in CAnnoFormatBlock::Deserialize , the function that processes text-formatting data between sharers and viewers. That function trusts remote participants’ character-length values without checking if they fit into its fixed 128-byte buffers. Sending an oversized annotation object overflows the buffer, corrupts adjacent memory, and hands an attacker control of program execution. The attacker’s entry requirement: join the meeting. That’s it. No social engineering, no malicious links, no user interaction on the victim’s side. A Security demonstrated the flaw on macOS by silently launching Safari on a target machine mid-meeting. Researchers could steal data, activate cameras and microphones, or install malware — all without generating any visual warning for the target. The same vulnerable code compiled across Windows, macOS, iOS, Android, and Linux simultaneously. Three other CVEs travel with the chain: CVE-2026-53414 buffer over-read, denial of service , CVE-2026-53415 use-after-free, memory corruption , and CVE-2026-53416 path traversal in the Zoom VDI Client . All patched August 11 in Zoom’s security bulletins ZSB-26015 through ZSB-26018 https://a.security/blog/asecurity-zoomsday . The AI Story Is the Real Headline The vulnerability matters. The method of finding it matters more. A Security’s researcher directed an AI agent at Zoom’s annotation component — recognizing that obscure functions in proprietary protocols make good targets. The AI independently reasoned about protocol structures, generated multiple attack vectors, and iterated on failures without human intervention at each step. What followed was a working, weaponized exploit. Time elapsed: under 24 hours. Prompts used: fewer than 20. This is work that traditionally required nation-state infrastructure: a five-person elite team, six months of effort, and resources measured in the millions. The firm was direct about what this means: “The barrier that kept these weapons scarce has collapsed, and it won’t come back.” One important nuance: the AI didn’t autonomously discover the bug. Automated scanning ranked the vulnerable function 45th in its initial output — a human had to point the AI at the right target. What AI eliminated was the months of manual exploitation work that followed. That distinction matters for how we evaluate the threat, but it doesn’t soften the conclusion. The timing wasn’t coincidental. OpenAI restricted its most advanced offensive-security model, GPT-5.6-Cyber, to vetted security partners the day before the Zoomsday disclosure. The company noted that gated versions answer 95% of advanced offensive-security prompts versus 1.5% for public models. Model access controls are a meaningful speedbump — not a solution. What Developers Need to Do The patches are out. This is now a deployment problem. Zoom Workplace : Update to 7.1.5 or 7.0.6 on all platforms Zoom Rooms : Update to 7.1.5 Meeting SDK : Update to 7.1.5 — critical if you’re embedding Zoom in your application VDI Client : 7.0.11 or 6.6.16; VDI Plugin: 7.0.11 or 6.6.15 For admins running meetings with external participants: disable annotation in your account settings if you don’t use it. The same logic applies to file transfer, remote control, and third-party apps. Attack surface you don’t use is attack surface you don’t need to defend. Require passcodes and waiting rooms for any meeting that isn’t fully controlled. For developers embedding the Meeting SDK: this is the one that often gets missed. SDK versions don’t update themselves, and they sit in your build dependencies. Check yours now https://www.malwarebytes.com/blog/bugs/2026/08/zoomsday-flaws-could-let-one-zoom-participant-attack-another . The Threat Model Has Changed The standard assumption in enterprise security has been that zero-day exploit development for major commercial platforms requires nation-state-level resources. That assumption is no longer safe to make. 28.3% of CVEs are already being exploited within 24 hours of public disclosure. AI-assisted research compresses the window defenders have between patch release and active exploitation. Zoomsday was responsibly disclosed: A Security waited until both a client-side patch and a server-side mitigation were deployed before going public. That gave Zoom and its users breathing room https://securityaffairs.com/197042/hacking/zoom-patches-zoomsday-zero-click-flaw-enabling-remote-code-execution.html . Not every future disclosure will be that patient. If your security posture assumes complex features in commercial software are low-priority targets because they’re hard to exploit, Zoomsday is a counter-argument worth taking seriously.