Zoom has patched three memory corruption flaws in its annotation feature that allowed any meeting participant to run code on another attendee’s device with no interaction. The fixes shipped in June and July 2026, roughly two months before the research was made public.

Zoom has patched three flaws in the annotation tools used during screen sharing that let anyone on a call run code on another participant’s device. No click was needed and nothing visible happened. The fixes shipped in June and July.

That timing is worth stating plainly, because the story has been written up as an emergency. Zoom closed the holes roughly two months before the research went public, so anyone on a current client is already covered.

The fixed builds are Zoom Workplace 7.1.5 and 7.0.6, Rooms and the Meeting SDK at 7.1.5, and the Windows VDI client at 7.0.11 and 6.6.16. Anything older remains exposed.

The severity is also softer than reported. A Security, the firm that found the bugs, scored all three at 9.0 out of 10. Zoom rates CVE-2026-53413 and CVE-2026-53415 at 8.3 and CVE-2026-53414 at 6.5.

The mechanics are ordinary memory-safety failures. Annotations cross the network as a run of counts followed by data, and the receiving client trusted those counts, letting one value overrun a 128-byte buffer and corrupt a return address. A second flaw let the dispatcher accept annotation messages without checking which participant had sent them.

An attacker on the call, host or guest, could send a crafted message that executed on other machines with no prompt and no indication. What follows is the usual menu, file theft, camera and microphone access, credential and wallet harvesting, second-stage payloads.

The headline claim is that AI did the work. A Security says it went from flaws to working exploit in under a day, using fewer than 20 prompts on publicly available models, and argues that this capability was until recently the preserve of nation states. OpenAI has meanwhile shipped a cyber model trained to refuse less.

Its own writeup complicates that. An automated ranking pass across 3,762 functions missed the vulnerable library entirely, placing it 45th, and the bug only surfaced when a researcher traced a live call by hand. AI weaponised it fast, but a human found it.

The wider shift is real regardless. Anthropic’s Mythos has found 10,000 critical vulnerabilities in a month, faster than anyone can patch them.

Get the TNW newsletter

Get the most important tech news in your inbox each week.