ProBackend
Vulnerabilities & Exploits
2 weeks ago5 min read

APT29's watering-hole weapon: zero-click exploits for flaws you already patched

APT29's watering-hole weapon: zero-click exploits for flaws you already patched

Sekoia's Threat Detection & Research team documented a large-scale watering-hole campaign it links to Russia-linked APT29, silently infecting iOS and Android visitors with exploit chains built entirely from already-patched n-day bugs. Here's how the three infection doors, the zero-click chain, and the CVE list actually fit together — and why the patch gap is the whole point.

A quiet infection that needs no click

The nastiest part of the campaign Sekoia's Threat Detection & Research team published in early November isn't the exploits. It's the silence. APT29 — the Russia-linked group also called Midnight Blizzard or Cozy Bear, tied to the SVR foreign intelligence service — poisoned legitimate websites, spun up convincing lookalike domains, and for about a week backdoored a popular analytics service, all so that a visiting iPhone or Samsung owner could be profiled and exploited without tapping a thing. No link, no attachment, no "please enable this." Sekoia spotted the operation in mid-October 2026 and assesses the attribution with moderate-to-high confidence.

You bring your own patch risk. They bring the rest.

Three doors, one payload

Most watering-hole operations pick a lane. This one used three delivery vectors at once, and Sekoia is explicit that all of them funnel into the same thing: a code injection that loads an external script to fingerprint the visitor's browser. The vectors are direct JavaScript injection onto legitimate compromised sites, adversary-in-the-middle lookalike domains that relay the real site's content so the URL and certificate look trustworthy, and a supply-chain compromise of Plerdy, a Ukrainian web analytics service.

The legitimate sites Sekoia identified are exactly the kind that attract the targets a state intelligence service actually wants. A Ukrainian gas producer. A U.S. chemical-industry trade association. A Bulgarian site with military ties. The lookalikes were even more pointed: one impersonated the official America250 site, the other the Raffles London at The OWO hotel. Then there's Plerdy. Between August 27 and September 3, thousands of websites — mostly in Ukraine, loaded its modified SEO tracker on their own visitors' behalf, which is what a supply-chain position buys you. You don't chase victims. You wait where they already go.

How the zero-click chain actually works

Fingerprinting comes first, and it's the quiet gate. The injected JavaScript reads the device and browser version, sends that to a command-and-control server, and waits. The C2 then picks the matching exploit chain from stock and serves it, no further interaction required. Sekoia's Bill Toulas coverage put it plainly: the victim's browser never has to do anything. This is zero-click in the way the term is supposed to mean, and note it leans on browser and OS attack surface rather than, say, iMessage or another messaging app, which is where zero-click usually gets its reputation.

So the real skill here isn't a fresh vulnerability. It's a decision: how few of your targets do you want to bother with?

The CVEs doing the work

The chains are all n-days, vulnerabilities that already had fixes sitting on the vendor's shelf. On the iOS side, the analyzed chains used CVE-2025-4320, an iOS kernel bug, and CVE-2025-31200, a flaw in the IOAcceleratorFamily driver that interfaces with the GPU. On the Android side, in a single attack Sekoia observed against Samsung phones, it used CVE-2025-21042 in Samsung's Exynos baseband chip and CVE-2025-30479 in the ARM Trusted Firmware-M firmware, the latter discovered by Google's Threat Intelligence Group and a Google arms-length entity.

One more CVE matters for context. CVE-2025-43529, a CoreGraphics type-confusion bug Apple patched back in August 2025, isn't in these chains, it belongs to the separate DarkSword exploit kit, where it was used by the Chaos ransomware crew and the UNC5221 spyware group to backdoor Chrome for iOS and phish people. Apple even backported that CoreGraphics fix down to iOS 18.7.2 and to legacy builds for 16.x and 15.x. Sekoia also found four additional vulnerabilities sitting in unreachable code or alternate exploit paths that the analyzed chains never executed. The takeaway from the whole list: most of the damage happens to people who simply never hit "update."

Persistence you can reboot away

Not everything about this operation is sophisticated, and that's revealing. Sekoia said none of the APT29 payloads established persistence. Reboot the device and the malicious activity stops, until the user wanders back onto a compromised site and the whole profile-and-pick-your-chain routine runs again. For an intelligence-collection mission aimed at grabbing cookies, messaging app databases and credentials (including Wi-Fi passwords), that's an acceptable trade. You don't need a foothold for months if the victim keeps returning to the same watering hole. It's a low-footprint model that trades durability for stealth.

Why it belongs in the AI cybersecurity threats conversation

State-sponsored actors increasingly lean on machine-speed targeting and automated exploit selection, and that's the thread worth pulling from this campaign into the broader discussion of AI cybersecurity threats. The C2 didn't hand every visitor the same package; it profiled the browser and chose a tailored chain. That decision loop is exactly the kind of targeting that AI-assisted tooling makes cheaper and faster for groups like APT29. None of this proves AI wrote the exploits, nothing in Sekoia's write-up claims that, but it shows the selection-and-delivery layer is already the bottleneck worth hardening, and it's the layer defenders should assume is getting smarter. The CISA guidance to patch and segment doesn't get less true here; it gets more urgent when the picking gets automated.

What the playbook tells defenders

APT29 isn't improvising. Over the summer Microsoft reported the same group spent months hijacking hotel and conference-center Wi-Fi to lift Microsoft 365 credentials, so mobile collection is a running theme, not a one-off. The pattern across both operations is the same: stand near where valuable people gather, exploit the slow to patch, keep a light footprint. Defensively, that points at the unglamorous controls, fast patch rollout, especially the backported fixes that keep older iPhones in the game, network segmentation for the un-updateable, and treating third-party analytics and chat widgets as supply-chain surface rather than decoration. If your threat model includes anyone with the SVR's patience, the patch gap isn't a hygiene checkbox. It's an open door, and right now APT29 is walking through it silently.

Source: BleepingComputer

a quiet infection that needs no clicka quiet infection that needs no click