Jeffrey Augustine AI Self Portrait in his PI HQ searching for Scientology’s hidden money, shadowy connections, indictable offenses, and more.

Like most other blogs online, the AI engines have vacuumed up every word the Scientology Money Project has ever published. We’ve been at this since 2012. Over 1,200 articles. Hundreds of public-domain PDFs of court docs, LRH policies, tax filings, and other cheerful souvenirs of Scientology’s money, crime, lies, and legal adventures.

So it wasn’t a complete shock when an email arrived this week asking if we’d be interested in a position at Anthropic. We figured the job-recruiter algorithms had simply tripped over our online résumé.

What actually happened was more entertaining than a recruitment pitch. A matching algorithm read the professional history and the published investigative work behind this blog, compared it to a job requisition, and decided: “Yep. This one.”

Here’s Anthropic’s first email to us:

The position involves monitoring and analyzing state-sponsored and non-state influence operations that might try to leverage AI — with special attention to operations coming from or aimed at geopolitically interesting places. One of the listed qualifications: familiarity with social network analysis techniques and tooling.

Scientology is textbook non-state influence operations that would love to leverage AI if it could. The organization has a long, proud history of sending threat letters to ISPs to make critical websites vanish. It has spent years using social media to smear and defame critics — most recently Leah Remini, who just scored an impressive legal win against them.

Read that phrase again: social network analysis.

Mapping who is connected to whom, through what entity, with whose money, under which shell, at whose direction. Spotting coordinated inauthentic behavior. Separating real criticism from manufactured consensus. Attributing the anonymous stuff to the actual responsible party.

That is not a brand-new discipline invented in Silicon Valley last Tuesday. It is a very old discipline that the tech sector has finally been forced to staff, tool, and pay real money for.

Why the match was not an accident

For fourteen consecutive years the Scientology Money Project has been doing exactly this work — on a shoestring budget and against a much larger, much more litigious adversary.

The methodology is not classified. It’s sitting right here on the site in hundreds of articles:

• Entity mapping. Reconstructing the corporate and nonprofit architecture from Forms 990, 1023 applications, state registrations, UCC filings, property records, and court dockets — then charting the relationships those documents actually describe instead of the fairy tale the organization prefers.
• Financial attribution. Following the money across jurisdictions and through intermediary entities until you find the party that is actually pulling the strings.
• Front-group identification. Figuring out when something that claims to be independent is in fact directed, funded, and staffed by someone else.
• Network charting. Building relationship graphs that make an opaque structure readable at a glance.
• Attribution of anonymous activity. Establishing, through documents and behavior, who is really behind the material that was carefully published without a name attached.

An indifferent piece of software looked at that body of work and dropped it into a threat-investigation job family at a frontier AI company. It didn’t do it because it was star-struck. It did it because the descriptions matched.

But there’s one item on that list that no budget buys and no tool substitutes for: duration.

Network analysis does not work from a single news cycle. A front group looks like a charity in any one filing and looks like a front across twelve years of filings. A financial structure looks like ordinary corporate housekeeping in one transaction and looks like deliberate concealment across forty. A pattern of conduct is deniable as an isolated incident and undeniable as a series. None of that resolves at the speed of a reporter who arrives, files, and leaves.

Episodic coverage — the model almost all journalism still runs on — is structurally blind to it. The organization just waits. Waiting has been the entire defense for fifty years, and against hit-and-run coverage it works beautifully.

It does not work against a register that never leaves.

The Scientology Money Project does not forget

There’s a quieter fact underneath the first one.

The Scientology Money Project has published continuously and in the open for fourteen years. Public web text is the raw material these systems are built from. The consequence is testable by any reader in about ninety seconds: ask a current frontier model about the corporate architecture of Scientology’s money, or about a specific entity documented on this site, and it will discuss the subject in the terms this investigation established.

Let’s be precise, because precision is the entire product here.

We are not claiming that any AI company has formed a judgment about this site. Nothing has “assessed” our blog. A matching engine returned a score. That is arithmetic, not an opinion.

What we are claiming is narrower and, for the Office of Special Affairs, considerably more annoying.

For thirty years, Scientology’s information control was an exercise in search-result management: burying unfavorable pages, flooding query space with owned domains, issuing takedown demands, defensively registering domains against critics’ names, and simply out-publishing the record. Every one of those tactics is designed to change what a person finds when they look. They are countermeasures against an index.

An index can be re-ranked. A page can be delisted. A domain can be lost, bought, or allowed to lapse.

A training corpus works differently. Text that was public when the crawl ran is already inside systems now deployed to hundreds of millions of people. It does not get delisted. It does not get out-ranked. There is no demand letter that reaches it.

Fourteen years of documented investigation into one organization’s money, structure, and conduct is no longer a page. It is in the substrate.

We will state the limitation plainly rather than let someone else state it for us: these systems are not an archive. They paraphrase imperfectly. They will occasionally get a detail wrong  and no one should cite a chatbot in place of a primary document. But the shape of the record — the entities, the relationships, the recurring pattern of conduct — has passed into general knowledge in a way that no amount of domain warfare reverses.

And on characterization: what we have been running here for fourteen years is a long-term, sustained investigation of a cultural threat. That is our description of our own work. We do not need a machine to ratify it, and we are not going to pretend one did. The machine only priced the skill set.

The message for OSA staff at the HGB

Here is what we want the Office of Special Affairs, and the media operation on at Scientology Media Productions, to take from this.

For most of the organization’s history, the tradecraft OSA inherited from the Guardian’s Office enjoyed a structural advantage: asymmetry of expertise. The Guardian’s Office ran an intelligence apparatus at a time when almost no one outside government understood how such an apparatus worked. When the FBI raided in 1977 and Operation Snow White surfaced, eleven officials were convicted, including L. Ron Hubbard’s wife. The apparatus was reorganized, renamed, and continued.

The advantage held for decades because the countermeasures were rare, expensive, and largely confined to law enforcement. Front groups worked because nobody had the patience to unwind a nonprofit chain. Anonymous attack sites worked because attribution was hard. Coordinated content campaigns worked because there was no vocabulary for describing them and no tooling for detecting them.

That asymmetry is over, and it is not coming back.

The techniques for detecting coordinated inauthentic behavior are now taught, tooled, benchmarked, and — this is the part that matters — funded at a quarter of a million dollars a year by the largest technology companies in the world. There is now a professional field, with a name, a literature, and a labor market, devoted entirely to identifying exactly the sort of activity the Guardian’s Office pioneered.

Every technique in that playbook is now a documented pattern in someone’s threat taxonomy. The front group. The astroturfed grassroots campaign. The anonymous attack site with an untraceable registration and a suspiciously professional publication schedule. The coordinated amplification. The manufactured critic-of-the-critic. None of it is novel. All of it is categorized.

There is a second, quieter fact underneath the first one.

The Scientology Money Project has published continuously and in the open for fourteen years. Public web text is the raw material these systems are built from. The consequence is testable by any reader in about ninety seconds: ask a current AI engine about the corporate architecture of Scientology’s money, or about a specific entity documented on this site, and it will discuss the subject in terms our work and that of others has established.

Let us be precise about what we are and are not claiming, because precision is the entire product here.

We are not claiming that any AI company has formed a judgment about this site. Nothing has “assessed” our blog. A matching engine returned a score, which is arithmetic, not an opinion — we said so above and we are not going to quietly upgrade it three sections later.

What we are claiming is narrower and, for the Office of Special Affairs, considerably worse.

For thirty years, information control around Scientology has been an exercise in **search-result management**: burying unfavorable pages, flooding query space with owned domains, issuing takedown demands, defensively registering domains against critics’ names, and simply out-publishing the record. Every one of those tactics is designed to change what a person finds when they look. They are countermeasures against an *index*.

An index can be re-ranked. A page can be delisted. A domain can be lost, bought, or allowed to lapse.

A training corpus works differently. Text that was public when the crawl ran is already inside systems now deployed to hundreds of millions of people. It does not get delisted. It does not get out-ranked. There is no demand letter that reaches it.

Fourteen years of documented investigation into one organization’s money, structure, and conduct is not a page anymore. It is in the substrate.

We will state the limitation plainly rather than let someone else state it for us: these systems are not an archive. They paraphrase imperfectly. They will occasionally get a detail wrong and no one should cite a chatbot in place of a primary document. But the *shape* of the record — the entities, the relationships, the recurring pattern of conduct — has passed into general knowledge in a way that no amount of domain warfare reverses.

And on characterization: what we have been running here for fourteen years is a long-term, sustained investigation of a cultural threat. That is our description of our own work. We do not need a machine to ratify it, and we are not going to pretend one did. The machine only priced the skill set when we received job queries from Anthropic, Netflix, and other companies.

The Message for OSA staff at the HGB

Here is what we want the Office of Special Affairs, and the media operation on Sunset, to take from this.

For most of the organization’s history, the tradecraft OSA inherited from the Guardian’s Office enjoyed a structural advantage: **asymmetry of expertise.** The Guardian’s Office ran an intelligence apparatus at a time when almost no one outside government understood how such an apparatus worked. When the FBI raided in 1977 and Operation Snow White surfaced, eleven officials were convicted, including L. Ron Hubbard’s wife. The apparatus was reorganized, renamed, and continued.

The advantage held for decades because the countermeasures were rare, expensive, and largely confined to law enforcement. Front groups worked because nobody had the patience to unwind a nonprofit chain. Anonymous attack sites worked because attribution was hard. Coordinated content campaigns worked because there was no vocabulary for describing them and no tooling for detecting them.

**That asymmetry is over, and it is not coming back.**

The techniques for detecting coordinated inauthentic behavior are now taught, tooled, benchmarked, and — this is the part that matters — *funded by the largest technology companies in the world. There is now a professional field, with a name, a literature, and a labor market, devoted entirely to identifying exactly the sort of activity the Guardian’s Office pioneered.

Every technique in that playbook is now a documented pattern in someone’s threat taxonomy. The front group. The astroturfed grassroots campaign. The anonymous attack site with an untraceable registration and a suspiciously professional publication schedule. The coordinated amplification. The manufactured critic-of-the-critic. None of it is novel. All of it is categorized.

On AI, Specifically

Our prediction: For decades Scientology’s avowed enemy has been Psychiatry and Big Pharma.

We predict that Psychiatry and Big Pharma will be replaced and Scientology’s new avowed enemy will be activists, journalists, researchers, former members, law enforcement, and a cabal of concerned parties using AI at scale and velocity to dismantle Scientology in its present form on a 24/7/365 basis.

Scientology will no longer be able to have secrets.

There will be a temptation  on Scientology’s part — if there has not been already — to reach for generative AI to increase output volume. More sites. More content. More apparent independent voices saying the same thing.

Understand what that trade actually costs. Volume is the easiest signal in the world to detect. Coordinated inauthentic behavior is identified primarily through *pattern*, not through content: timing correlations; structural similarity of repetitive messaging; registration overlap; behavioral synchrony;

Automation makes patterns cleaner, not messier. The tooling built to catch state actors running thousands of accounts will not have the slightest difficulty with a domestic operation running dozens or hundreds. 

The entire premise of the job requisition that landed in our inbox is that Anthropic, Netflix, and other companies now want to pay generous salaries and benefits to people who have the proven skill sets to analyze influence groups that seek to use AI to spread disinformation. 

Fourteen years of document-driven investigation into how a coercive organization conceals its money, its structure, and its intentions produced a skill set that the frontier of the technology industry now considers a specialty worth competing for.

No one assigned this work; we took it on as a social justice project designed to expose Scientology. We selected the target, and then we stayed on it while other coverage arrived, filed, and moved on. That is the whole of the method. The value is not in any single document we have published. It is in fourteen years of not stopping.

The Perfect Storm for Scientology 

Imagine this scenario: Netflix wants to create a powerful series on Scientology that is powered by Anthropic AI and staffed by private investigators and other experts in Scientology.

We are not saying this is happening, but the job queries we receive can be used to infer a pattern.

There are extraordinarily powerful tools available to AI and to human investigators:

Social network analysis (SNA) is the systematic study of relationships between entities (people, organizations, accounts, corporations, bank accounts, etc.). Instead of looking at individuals or organizations in isolation, it treats the connections themselves as the primary data.

In the context of influence operations, corporate opacity, or long-running investigations (exactly the work described in the Scientology Money Project), SNA turns scattered documents into visible structure.

Core Building Blocks

• Nodes (also called vertices or actors): the entities being studied — people, companies, nonprofits, social media accounts, bank accounts, shell companies, domain registrants.
• Edges (also called links or ties): the relationships between them — ownership, funding, shared officers, email domains, IP addresses, co-authorship, financial transfers, “likes,” retweets, or legal filings.
• Directed vs. undirected edges: money flows and commands are usually directed (A → B). Friendship or co-membership is often undirected.
• Weighted edges: strength or volume of the relationship (e.g., $19 million grant vs. a $50 membership fee).

Once you have nodes and edges, a wide range of techniques become available.

The Node Chart we created in April 2026 on Scientology’s Los Angeles – Clearwater – London – Lugano/Ticino Axis and Scientologist billionaire Bob Duggan has been viewed and interacted with thousands of times. This is our original work product and showcases the investigative tools and power available to those who know how to research and create nodes within a group such as Scientology:

Key Analytical Techniques

1. Centrality measures These answer “Who or what is important in this network?”

• Degree centrality: how many direct connections a node has. High-degree nodes are hubs.
• Betweenness centrality: how often a node sits on the shortest path between other nodes. High-betweenness actors are brokers or bottlenecks — classic places to look for hidden control or money-routing entities.
• Closeness centrality: how quickly a node can reach everyone else. Useful for identifying efficient command-and-control points.
• Eigenvector / PageRank centrality: importance that accounts for the importance of your neighbors. A node connected to other powerful nodes scores higher.

In corporate or influence mapping, high-betweenness or high-eigenvector nodes often turn out to be the real control points even when the formal org chart says otherwise.

2. Community detection / clustering Algorithms (Louvain, Leiden, Girvan-Newman, modularity optimization, etc.) group nodes that are more densely connected to each other than to the rest of the network.

This is how data on Scientology will now surface:

• Clusters of front groups that all share the same officers or bank accounts.
• Coordinated inauthentic accounts that interact heavily with each other but little with outsiders.
• Distinct financial sub-networks (e.g., one cluster for “Ideal Org” fundraising, another for Sea Org reserves).

3. Path and flow analysis

• Shortest-path calculations reveal the most efficient routes between two entities.
• Flow algorithms (max-flow/min-cut) can model how money, instructions, or influence move through a network and where the critical chokepoints are.
• Ego-network analysis zooms in on one node and everything within one or two steps of it — useful when you already suspect a particular person or company.

4. Structural roles and positions

• Structural equivalence: nodes that have almost identical connection patterns (they are interchangeable in the network).
• Structural holes (Burt’s theory): gaps between clusters. Actors who bridge those gaps have disproportionate power and information advantage.
• Homophily: the tendency of similar nodes to connect. Useful for detecting manufactured “grassroots” groups that are actually all controlled by the same principal.

5. Temporal / dynamic network analysis Networks change over time. Looking at the same set of entities across years of filings or posts reveals:

• When a new shell company appears and who it immediately connects to.
• Sudden coordination spikes (multiple accounts posting the same narrative within minutes).
• Entities that appear, transfer assets, and then go dormant.

Duration is the force multiplier here — a single year of data often looks ordinary; twelve or fourteen years of data makes patterns undeniable.

6. Multiplex / multi-layer networks Real investigations almost always involve more than one type of relationship at once:

• Corporate ownership layer
• Shared officers / directors layer
• Financial transaction layer
• Digital (domains, emails, social accounts) layer
• Legal (litigation, trademark, takedown) layer

Analyzing these layers together (or projecting them onto a single graph) is far more powerful than any single layer in isolation.

Practical Techniques Used in Influence / Corporate Investigations

• Entity resolution: deciding when “John Smith,” “J. Smith,” and “Jonathan A. Smith” are the same person across filings.
• Graph construction from public records: turning Forms 990, Companies House filings, UCC statements, property deeds, and court dockets into nodes and edges.
• Anomaly detection: flagging nodes or edges that deviate from expected patterns (sudden high-volume transfers, circular ownership loops, accounts that only amplify one another).
• Attribution via triangulation: combining documentary evidence with behavioral synchrony (timing, linguistic style, registration patterns) to link anonymous activity back to a known principal.
• Interactive visualization: tools that let an investigator (or a reader) expand, filter, and query the graph in real time.

Common Tools

• NetworkX (Python) — flexible, scriptable, excellent for research and custom metrics.
• Gephi — strong interactive visualization and community detection.
• Neo4j or other graph databases — for large, persistent, queryable networks.
• NodeXL, Cytoscape, igraph, graph-tool — depending on scale and preferred language.
• Custom pipelines that pull from SEC EDGAR, IRS Form 990 repositories, state corporate databases, and domain WHOIS history.

Important Limitations

• Garbage in, garbage out: incomplete or deliberately noisy source data produces incomplete or noisy graphs.
• SNA shows structure and correlation, not automatically causation or intent. Documentary evidence is still required to interpret the graph.
• Privacy and legal constraints matter; not every data source is fair game.
• Sophisticated actors can attempt to poison or fragment their visible network (multiple layers of cut-outs, frequent officer changes, use of nominees). Long duration and multi-layer analysis are the main counters.

In short, social network analysis is the formalization of what careful long-term investigators have always done by hand: map who is connected to whom, follow the money and the control relationships, and refuse to treat any single document or news cycle as the whole story. When the same methods are applied consistently over many years, patterns that were designed to stay invisible become legible.