Pegasus Took a Phone’s Past. AI Makes It Attributable

Key Takeaways

What does NSO Group’s own Pegasus product description say the spyware could collect, and why does it matter for attribution?

NSO’s August 2018 Pegasus 3.0 product description, now filed in Meta’s WhatsApp lawsuit, pitches zero-click installation, live interception, and “historical data extraction” of a phone’s messages, calls, contacts, browsing, and Wi-Fi history. That is today. Tomorrow, AI and ever-growing data streams make the same retrospective data the strongest tool for attributing operators and for vetting people.

Pegasus is NSO Group’s mobile spyware: software that, once planted on a phone, collects what the phone holds and what it hears. For years the public picture of it came from forensic researchers, leaks, and reporters. Now a version of the company’s own sales pitch sits on a federal court docket.

The document is titled “Pegasus Version 3.0 — Product Description” and dated August 2018. It was filed on September 15, 2025, as Exhibit 32 in WhatsApp v. NSO Group, the lawsuit Meta’s messaging service brought against NSO in the U.S. District Court for the Northern District of California. OctoProbe hosts the full sixty-seven-page filing below.

This op-ed reads the document for one capability above all: harvesting a phone’s past. Every capability described here is NSO’s claim, made to prospective government buyers, not independently verified fact. Then the piece asks the forward question. If an implant could pull a target’s history in minutes in 2018, what does artificial intelligence do to the value of history for attribution and for vetting?

What Is the Pegasus 3.0 Product Description?

It is NSO’s own overview of Pegasus for government customers, produced in discovery in WhatsApp’s lawsuit. It fills pages 2 through 36 of Document 796-15, stamped “Highly Confidential – Attorneys Eyes Only,” with Bates numbers beginning NSO_WHATSAPP_00045591. Its confidentiality page limits its use to buyers weighing a Pegasus license.

That page says the document may be used only for “evaluating the possibility of acquiring a license to use one or more of our solutions” or, after a purchase, for product support and training. The pages carry the brand of Q Cyber Technologies, and the copyright line names “Q Cyber Technologies SARL and its affiliates.” Q Cyber Technologies Limited is NSO’s co-defendant in the case.

The filing holds more than one exhibit. Exhibit 33, on pages 37 through 59, is a second NSO-produced document, a “Mobile Endpoint Product Description” with no date on its face. It refers to “the Company” and “the mobile endpoint solution” rather than naming Pegasus. Exhibit 35, on pages 60 through 67, is a hardware bill of materials for the customer site: Dell servers, NetApp storage, Cisco routers, and software licenses.

Each exhibit’s cover sheet reads “Unredacted Version of Document Proposed to Be Filed Under Seal.” Passages throughout are blacked out and stamped “Redacted – Export Controlled.” This article quotes around those passages and does not guess at what they hide.

The case has moved on since NSO wrote the pitch. The final judgment lays out the sequence.

What Did NSO Say Pegasus Could Do?

According to the document, Pegasus plants a hidden software “agent” on a target’s phone, then collects messages, calls, locations, files, and passwords and can switch on the microphone and camera. NSO says installation needs only a phone number and, in its covert mode, no click at all. The unredacted text names Android and BlackBerry; another platform is blacked out.

A few terms first. An implant is software secretly installed on a device to collect from it; NSO calls it an agent. A zero-click installation compromises a phone without the target tapping anything. An MNO is a mobile network operator, the carrier. C&C means command and control, the operator’s servers that send orders to the implant and receive what it collects.

NSO’s covert method is zero-click by its own description. “A covert message is sent to a mobile device,” the document says, and “the target is not engaged—there is no need to click a link or open a message.” The same page says “only a target’s phone number (MSISDN) is required for a successful installation.” A second method, which NSO calls “Triggered (Social Engineering),” relies on a crafted text, instant message, or email: “One click, whether active or unintentional, leads to an agent installation.”

Section 2.1, “Benefits of Pegasus,” lists eleven claims. One promises to “Remotely and covertly collect information” about “a target’s relationships, locations, phone calls, plans, and activities.” Another promises surveillance “regardless of whether or not a target switches between virtual identities and SIM cards.” A third says, “Cooperation with local MNO is not required.”

The table below collects the headline claims with their pages in the filing.

The last claim deserves a caution. “No trace” is a sales line. In July 2021 Amnesty International’s Security Lab released the open-source Mobile Verification Toolkit, built to gather “forensic traces helpful to identify a potential compromise of Android and iOS devices,” together with a forensic methodology for Pegasus. The marketing and the forensics do not agree.

What Is Historical Data Extraction?

Historical data extraction is NSO’s term for pulling everything already stored on a phone when the implant lands, not just what arrives afterward. The document lists SMS records, contacts, call history, calendar records, emails, instant messages, browsing history, installed applications, and Wi-Fi networks, all sent to the operator’s C&C servers.

NSO framed it as the product’s edge over rivals: “Pegasus enables the extraction of all data from a device—existing intel and real-time additions.” It added, “Other market offerings are limited to data generated after the installation and partial monitoring.”

Then came the line that sums up the pitch: “Normally organizations devote months to collecting such information, with Pegasus this is achieved in minutes.” NSO added that the history helps “build a comprehensive and accurate intelligence picture of a target/s and their associates.” A note says the feature can be turned off if an organization “is not permitted to access and extract historical data.”

The companion document in Exhibit 33 goes further. Its section 4.4, “Request Historical Data,” says the operator can order the implant to “collect all the historical data from the target’s mobile device” during installation “or later on at any given time.” It says the implant will “extract all the information that exists in the mobile device’s storage from the beginning of time until the current moment.”

The supported types run from call logs and instant messaging to “Connectivity history” and “File lists and notes.” Elsewhere the documents push the timeline back further still. The Pegasus description says, in a passage on cloud data that is partly redacted, that “Data stored on mobile devices merely hints at the mass of available historical information,” information “dating back months and even years.”

The analysis tools are built for the same purpose. Exhibit 33 describes a date filter that lets an operator “take a look at the earliest activities found on the target’s device.” A “Data History” view tracks changes such as an updated contact or a new Wi-Fi password to “unveil the secrets even the most sophisticated targets are trying to hide.”

The commercial logic is plain. A target can grow careful today. A target cannot go back and un-send last year’s messages, un-join last year’s networks, or un-visit last year’s websites. NSO priced the past as the part of a person that caution cannot reach.

How Did NSO Say It Hid the Operator?

The document describes an “Anonymizing Transmission Network,” relay nodes spread worldwide that send each implant’s traffic along varying paths before it reaches the customer’s servers. NSO wrote that the network makes it “impossible to trace back to an operating organization thus ensuring full deniability.” That is signature reduction by design.

Signature reduction means shrinking the traces an operator leaves for others to find: network paths, infrastructure, timing, and evidence on the target’s device. The section on transmission says communication runs “via an anonymizing network—thus trace-back is infeasible.” The same page says the implant goes dormant in a “country-of-risk,” such as one with high exposure risk or costly roaming, and holds its data until the target leaves.

Section 13 draws the setup. Target devices connect by cellular, Wi-Fi, or direct links into public networks. A wide-area network of relay nodes, one labeled “Anonymizing Node,” carries the traffic to a “Customer Site” of cellular modems, an Asterisk PBX phone server, storage, database, web, and management servers, and operator terminals. NSO says the “Anonymized nodes serve only a single client” and that all servers sit “inside the client’s trusted network.”

The device side gets the same treatment. The implant watches for “connections to forensic tools” and can self-destruct if exposure looks likely. If it loses contact with its servers, it removes itself after a default period of twenty-one days.

The same system also keeps meticulous history of its own. Collected data is timestamped “At creation” and “Upon arrival at the C&C servers.” Audit records of every user action “cannot be deleted or altered by any level of system user.” In Exhibit 33, “Every activity that is performed by the target and is therefore extracted by the endpoint, has a location attached.” The product reduced its operators’ signature on the wire while building a durable record on its targets and on its own users.

Why Is Attribution Getting Faster?

Attribution, tying an operation to the person, unit, or government behind it, has long been a race between an operator’s caution and an analyst’s patience. AI shortens the analyst’s side. Correlating years of logs, device records, and infrastructure traces once took a team months. It is becoming a machine task measured in minutes, the same compression NSO sold.

That is today. The question is what happens tomorrow.

AI-driven cyber activity is accelerating on both sides. Offense writes, tests, and adapts tooling faster. Defense triages alerts, clusters infrastructure, and matches indicators faster. When correlation gets cheap, the limiting factor stops being analyst hours. It becomes the depth and reach of the data, and above all the historical data.

Public Pegasus forensics already work that way. Researchers do not wait for an implant to announce itself. They examine records a phone already holds and match them against published indicators. The method looks backward by design. AI makes that backward look wider, faster, and cheaper.

Can Historical Data Trace Actors in Hard-to-Reach Jurisdictions?

Increasingly, yes, in the author’s view. An operator in a country beyond subpoena or extradition still leaves history outside that country: on targets’ devices, foreign infrastructure, commercial data sets, breached databases, archived web pages, and the logs of every service touched. Those records outlive the operation and can be correlated later.

The Pegasus pitch, reversed, explains why. NSO’s selling point was that the past stored on one device beats months of live collection. Attribution works the same way. An operator’s past is reused infrastructure, persona accounts, working hours, travel, writing habits, and tooling choices. It sits in many jurisdictions at once, most of them not the operator’s own.

Today, attribution against an actor in a hard-to-reach country often depends on what defenders catch live. Tomorrow it will lean on backfilled history: data collected for other reasons, years earlier, and joined to the case only when a machine finds the link. A jurisdiction can shield a person from arrest. It cannot recall every record that person generated abroad.

That raises the bar for rigor, not just the speed. Historical data can be wrong, stale, or planted as a false flag. Machine correlation produces candidates, not proof. Attribution still needs independent corroboration and a clear account of what each record does and does not show.

Why Does Signature Reduction Get Harder Over Time?

Because the data never stops piling up. Signature reduction works on what an operator controls now: new infrastructure, clean devices, and disciplined habits. It cannot reach back and scrub what phones, networks, sensors, and services already recorded. Each year adds more connected devices and data streams, so the record an analyst can mine grows faster than an operator can hide from it.

NSO saw the problem from the collector’s chair. Exhibit 33 warns that “Mobile devices store tens of thousands of records and generate vast amounts of data by the minute,” and calls navigating that “sea of data” a daily challenge for the analyst. The same page concedes that “Today’s world has become an increasingly monitored world with the vast internet traffic being monitored by web services and governments.”

In 2018 volume was the analyst’s burden. With AI, volume becomes the analyst’s asset. Phones, cars, wearables, cameras, smart-home devices, and cloud services all generate logs. Each stream adds to an ocean of data that only grows over time.

An operator can still shrink today’s footprint. Tomorrow, the volume of historical data and the AI that correlates it make staying invisible across time much harder.

What Does This Mean for Security Clearances and Vetting?

Vetting is a backward-looking process, and it will deepen the same way. Today a Tier 5 investigation for Top Secret access carries a seven-year scope, a fixed lookback window that works much like a statute of limitations. With persistent historical datasets and AI, investigators can reconstruct a subject’s behavior across a whole life, not one window.

OctoProbe’s earlier report, 25 Pages of Press Clips: Leak Shows How Thin White House Vetting Really Was, found that 2016 White House political vetting ran on newspaper clippings. It also laid out the deeper federal system. Tier 5 replaced the Single Scope Background Investigation (SSBI) in fiscal year 2017 and starts with the Standard Form 86 questionnaire. The FBI’s deepest investigation level, by contrast, reaches the “Entire adult life.” That report argued that real vetting pulls dockets, deeds, filings, and breach data, not clips.

The Pegasus documents add the time dimension. If one device’s history could be pulled in minutes in 2018, a lookback window stops being a technical limit. It becomes a policy choice. Breach corpora, archived posts, court and property records, and device backups already persist for decades. AI can join them across a whole life.

That cuts two ways. Vetting can get far deeper, which the earlier report argued it needs. It also raises hard questions about fairness and context, because old records are noisy, and a twenty-year-old data point can mislead as easily as it can inform.

The adversary gets the same tools. The earlier report warned that adversaries mine breach dumps for leverage over officials. An implant with historical extraction on a cleared person’s phone delivers exactly that history. Vetting that stops at seven years may miss what a foreign service can already reconstruct.

Bottom Line

NSO’s 2018 pitch was about the past. Pegasus, as the company described it, did more than listen live. It pulled a phone’s history, “from the beginning of time,” in minutes, while routing its operators through relays built for deniability.

That is today. Tomorrow, AI turns history into the decisive layer of attribution and vetting. Operators can reduce their signature going forward. They cannot reduce the record they have already left, and that record keeps growing. The same is true of every person who will ever sit for a background investigation.

FAQ

What is the Pegasus Version 3.0 Product Description? It is an NSO Group sales and technical overview dated August 2018 and written for government agencies evaluating a Pegasus license. NSO produced it in discovery in WhatsApp v. NSO Group, and it was filed on September 15, 2025, as Exhibit 32 within Document 796-15 in the Northern District of California. Some passages are marked “Redacted – Export Controlled.”

What is historical data extraction in Pegasus? It is NSO’s term for pulling the data already stored on a phone when the implant lands, not just what arrives afterward. The document lists SMS records, contacts, call history, calendar records, emails, instant messages, browsing history, installed applications, and Wi-Fi networks. NSO wrote that organizations normally spend months collecting such information and that Pegasus does it in minutes.

What does zero-click mean? A zero-click installation compromises a phone without the target tapping a link or opening a message. NSO’s document describes a covert method in which “the target is not engaged” and says only the target’s phone number is required. Those are NSO’s claims in a marketing document, not independently verified findings.

Did NSO claim Pegasus leaves no trace? Yes. The document says Pegasus “Leaves no trace whatsoever on a target’s mobile device.” Forensic researchers later contested that in public. In July 2021 Amnesty International’s Security Lab released the open-source Mobile Verification Toolkit, which scans phones for forensic traces of spyware, including Pegasus.

How does AI change attribution? Attribution means tying an operation to the person, unit, or government behind it. AI makes correlation across years of logs, device records, and infrastructure traces fast and cheap. That shifts the limiting factor from analyst time to the depth of the historical record, which keeps growing and which operators cannot scrub after the fact.

How could this change security clearance background investigations? Today a Tier 5 investigation for Top Secret access carries a seven-year scope. Persistent historical datasets and AI make it practical to reconstruct a subject’s behavior across a whole life. That deepens vetting and raises fairness questions about old, context-free data. Foreign services can run the same kind of reconstruction against cleared people.

Colin Aamot is an investigative journalist and OSINT domain expert specializing in FOIA and public records. A technologist, U.S. Army veteran, and former Capitol Hill staffer, he has held director roles across the private and public sectors.Full bio