Pegasus: The Spyware You Can’t See Has Been Deployed Again

He didn’t click on anything. He didn’t download anything. He didn’t open any messages. Yet the student had spyware on his iPhone that could turn on his microphone, camera, and read his encrypted messages. Citizen Lab this week confirmed that NSO Group’s Pegasus was deployed via an iMessage zero-click exploit.

A student in Belgrade received a notification from Apple on his iPhone. He hadn’t clicked anything. He hadn’t downloaded anything. He hadn’t opened any suspicious message. Yet his phone was carrying spyware capable of switching on the microphone and camera and reading his encrypted messages. That spyware is called Pegasus.

Today we’ll walk through exactly what happened, what Pegasus and the company NSO Group are, how a zero-click attack works, why Pegasus on an iPhone lives only in memory and a reboot wipes it, how Android differs from iOS, who buys Pegasus, and whether there is any evidence of it being used in the Czech Republic and Slovakia.

What happened in Serbia

Let’s start with the report itself. On 2 September, Citizen Lab at the University of Toronto, working with Serbia’s SHARE Foundation, published a forensic analysis of an iPhone belonging to a member of the Serbian student protest movement.

He had received what Apple calls a Threat Notification — a warning that he had been targeted by mercenary spyware. I covered these notifications in one of the earlier Security Sunday episodes. The student contacted the SHARE Foundation, which passed the forensic artefacts on to Citizen Lab.

Analysis of the phone confirmed that the infection was carried out via an iMessage zero-click exploit and that the payload was NSO Group’s Pegasus. The infection indicators point, with high confidence, to the period between December 2025 and January 2026.

According to Citizen Lab, Apple had patched this exploit in iOS 18.4.1. That release came out in April 2025. The infection happened in December. In other words, the fix had existed for roughly eight months before the phone was compromised. We’ll come back to this, because it’s the business takeaway of the whole story.

The SHARE Foundation confirmed that since the beginning of 2026, at least 14 people in Serbia have been targeted with advanced spyware. Students, activists, a member of parliament and a local councillor. All from the opposition. And the timing lines up with the local elections held on 29 March. Twelve of them approached SHARE’s forensic team in August after receiving Apple’s warning. SHARE uncovered the other two cases through its own analysis, because they involved a new version of the Android spyware NoviSpy.

What Pegasus can do

So what can Pegasus actually do? Put simply: the attacker can do anything the user can do. Read notes, photos, and yes — encrypted messages. Plus covertly switch on the microphone and camera. End-to-end encryption won’t protect you from Pegasus. Signal, WhatsApp and Threema use encryption that protects a message in transit between you and the person you’re chatting with. Pegasus, however, runs beneath the application layer, with system privileges. It reads the message the moment the app decrypts it and renders it on screen.

How zero-click works

And what exactly is zero-click? Classic phishing needs you to click a link or open an attachment. A zero-click exploit needs nothing. It abuses a bug in an application that automatically processes incoming data before you ever see the message.

That makes iMessage an ideal target. When a message arrives, the iPhone parses attachments, generates previews and processes fonts and images in the background. Every one of those steps is a potential attack vector. The attacker sends a specially crafted message and the phone opens it on its own, with no action from the user. Hence zero-click.

Android is not safe either

If you think this is purely an iOS problem and you’re fine with an Android phone, let me correct that. On Android it’s even worse. Google named the Android version of Pegasus Chrysaor. Its capabilities are practically identical to the iOS version; what differs is the path the spyware takes to get in.

Samples analysed in 2016 and 2017 didn’t use zero-day vulnerabilities. They used the well-known Framaroot rooting technique, and when that failed, the app simply asked the user for permissions.

And there’s one more difference. While Pegasus on iOS lives in RAM and a reboot removes it, historical Android versions installed themselves into the system partition. That means they survived even a factory reset. They disabled system updates and deleted push messages to hamper forensic analysis. They could even remove themselves.

Amnesty reports that it finds more forensic traces on iPhones than on Android. Detection is easier on iOS because it’s a closed ecosystem with uniform, rich system logs. Android is fragmented. Different manufacturers, different versions, different logs. It’s easier to hide in the system.

Who buys Pegasus

So who can actually buy Pegasus? NSO sells exclusively to governments and government agencies. Every sale requires export approval from the Israeli Ministry of Defense. According to NSO’s own statements, Pegasus has been sold to at least 14 EU countries. So is Pegasus in the Czech Republic and Slovakia too?

In September 2024, Denník N reported that, according to four sources with ties to the security community, the SIS and other agencies had newly acquired spyware capable of taking over mobile phones, and that it was most likely Pegasus. However, ARTICLE 19 noted in September 2025 that the Slovak government had neither officially confirmed nor denied using Pegasus. So we have strong journalistic indications of a purchase. But no public forensic analysis from Citizen Lab or Amnesty confirming Pegasus on a specific Slovak phone.

And the Czech Republic? One piece of evidence exists. A court document from the WhatsApp v. NSO case shows that a 2019 campaign hit 1,223 users in 51 countries, including the Czech Republic. Who the victim was and which state was the operator, we don’t know.

What this means for businesses

How does all this translate to a corporate environment? Most companies have XDR on laptops, monitoring on servers and a firewall on the network. And the CEO’s iPhone? No MDM and no enforced updates. Remember Serbia. The exploit was patched eight months before the infection. Updating mobile devices is not optional.

So, first: update. Citizen Lab explicitly states that the zero-click used in Serbia was neutralised by the patch in the update.

Second: enable Lockdown Mode on iPhone. In March 2026, Apple stated it was not aware of any successful mercenary spyware attack against a device with Lockdown Mode enabled. The price is reduced functionality, but for executives and people with sensitive access, it’s a price worth paying.

Third: on Android, enable Google Advanced Protection.

And fourth: MDM with enforced patching.

Pegasus is not a Hollywood myth. In 2026 it’s running, it’s being sold, and it’s infecting unpatched phones. Encryption won’t save you. Updates and Lockdown Mode will — at least against what is known today.

TP-Link Patches Two Vulnerabilities in Archer AX55 V4

On 3 September, TP-Link published a security advisory describing two vulnerabilities in the Archer AX55 router, hardware version V4.

The first, CVE-2026–18167, is a stack-based buffer overflow (CWE-121) in the EasyMesh module, rated 7.7 / High under CVSS v4.0.

The second, CVE-2026–18330, involves the use of a hard-coded cryptographic key (CWE-321) in the web login module and is rated 6.1 / Medium.

Headlines along the lines of “vulnerabilities allow attackers to execute remote code” deserve to be qualified against the CVSS vector published by the vendor itself. CVE-2026–18167 carries AV:A — the attacker must already be connected to the router’s LAN or Wi-Fi, not coming in from the internet. The vector also includes AC:H (high attack complexity) and AT:P (specific preconditions must be met). For the EasyMesh daemon to be reachable at all, Mesh mode must be enabled.

CVE-2026–18330 is, in my view, the more interesting one. TP-Link states that the device contains a shared RSA-1024 private key. If the key is common to the entire product line, extracting it once from a single device is enough — it then applies to every other unit running the same firmware.

Both flaws are fixed in firmware 1.2.1 Build 20260527.

Dormant SQL Payload in All-in-One WP Migration and Backup

The Wordfence team has disclosed a high-severity vulnerability in the All-in-One WP Migration and Backup plugin. Tracked as CVE-2026–19949 and rated CVSS 8.8 (High), it affects all versions up to and including 7.109 and stems from a combination of insufficient escaping of user-supplied input and a missing prepared statement in the archive restore path (CWE-89).

ServMask, acting as its own CNA, published the record on 25 August 2026. Wordfence’s public technical analysis followed on 1 September.

According to WordPress[.]org data, approximately 3.2 to 3.25 million sites are running a vulnerable release.

Administrators of WordPress sites using this popular plugin should update to version 7.110 immediately, or deactivate the plugin if that isn’t possible.

If you restored a site on a vulnerable version any time since mid-August, treat ai1wm_secret_key as compromised, invalidate it and have a new one generated. Also check whether any unknown plugins or user accounts have appeared in WP.

AI Agents Took Over a Company in Under 10 Hours

Palo Alto Networks’ Unit 42 team has published an incident response case study describing a corporate network intrusion in which the human attacker handed almost all tactical execution over to AI agents and obtained root-level access in under ten hours.

By Unit 42’s estimate, a coordinated human red team would have needed roughly two weeks to achieve the same result.

Unit 42 explicitly states that no zero-day vulnerability was exploited and that the attacker demonstrated no exceptionally advanced techniques. The decisive factor was operational efficiency. The agents processing tool output decided on the next step and acted without waiting for a human to read the screen.

Unit 42 expects attackers to add AI agents to their toolkits with increasing frequency. Don’t expect sophistication, though — expect speed and the ability to parallelise tasks.