About

About this project

The Digital Harm Project is a research synthesis for parents, clinicians, educators, technologists, and policymakers who need a single grounded overview of a fast-moving evidence base. It takes no religious or partisan position. This page covers what the project is and how to cite it — and the editorial standards that govern every page.

Scope

The research covers eight subject areas: early pornography exposure among minors; neurobiological and clinical models of pornography addiction; the contested escalation pathway from legal pornography to child sexual abuse material (CSAM); the emergence of AI-generated pornography and CSAM; therapeutic interventions for both populations; consumer and industrial technology solutions; legislative prevention strategies; and consolidated recommendations for individuals, families, technology companies, policymakers, educators, and clinicians.

Citations and method

Every empirical claim is cited. The full citations list is available on the Citations page. Source types include peer-reviewed journal articles (preferred wherever available), government and NGO statistical releases, clinical reports, and journalism — generally in that order of preference. Where competing evidence exists (most notably around the gateway hypothesis from pornography to CSAM), the report presents the disagreement rather than resolving it.

We make a particular effort to flag data-integrity issues that have shaped public discourse, such as the Stanford Center for Internet and Society's 2026 finding that inflated “Generative AI” CSAM report figures arose from a categorization artifact in NCMEC's CyberTipline pipeline. Good policy depends on good numbers.

What this is not

This is not a medical resource. It does not provide diagnosis, treatment recommendations for an individual, or legal advice. If you or someone you know is in crisis, the Get Help page lists 24/7 services.

It is also not a comprehensive policy proposal. Where the research recommends specific policies (for example, privacy-preserving age-verification systems modeled on France's double-anonymity approach), those recommendations follow from the evidence presented; they are not the work of a legislative team.

Citing and reuse

You are welcome to cite, quote, and link to this research. Each chapter and section has a stable URL designed for direct citation. We ask only that you preserve linkbacks to primary sources rather than to our summary.

Timeline

The research underlying this project began in August 2023 and has been updated continuously since. The current edition was last reviewed in May 2026, with a substantial expansion adding ~150 new citations and new sections on CSAM production typology, distribution and economy, the reporting pipeline, detection technical depth, cross-jurisdictional law, the encryption debate, and survivors of CSAM circulation. The fast-moving areas in this research — AI-generated content, age verification law, and detection technology — will require periodic re-review.

The standards behind every page

The promise

A resource on this subject is only worth as much as it is trustworthy. The commitments below are our contract with you; if we ever fall short of them, hold us to it. Four govern everything here:

Accurate. Claims trace to a primary source. Where we summarize, we link to the original so you can check us.
Honest about uncertainty. When the evidence is mixed, we say so and show both sides rather than flattening it into a tidy answer.
Survivor- and person-centered. Our language follows the standards set by the people who work with survivors, not sensational convention.
Current. Every reference page carries a "last reviewed" date, and fast-moving statistics are re-checked against their sources.

How we source

We work outward from the most authoritative material available:

Primary sources first. Statutes link to the bill or the U.S. Code; statistics link to the issuing body's own report (NCMEC, IWF, WeProtect, Childlight, INHOPE); clinical claims link to peer-reviewed research or to bodies like the WHO and AASECT.
Named, dated, and quantified. A statistic without a year and a source is not a fact, it's a rumor. We attach all three or we leave it out.
We distinguish a measurement from an estimate. "Reports to the CyberTipline" is a count; "children affected globally" is a modeled estimate. We don't let the second masquerade as the first.
We note interpretation caveats. For example, a report flagged "generative AI" in NCMEC data is not the same as confirmed AI-generated CSAM — a distinction researchers at Stanford have stressed, and one we carry on the relevant pages.

The complete reference list lives on the citations page.

Terminology

Words shape how a harm is understood, so we follow the conventions established by survivors and the organizations that serve them:

"Child sexual abuse material" (CSAM), not "child pornography." The older phrase implies consent and a category of legitimate "pornography," both of which are false. This follows the Luxembourg Guidelines, the international standard on this terminology.
"Survivor" and person-first language. A person is not their worst experience or their diagnosis.
"Attraction" is distinguished from "offending." Conflating an unchosen interest with a chosen crime is both inaccurate and an obstacle to the prevention work that protects children.

How we handle the science

We are research-backed, and we follow the authoritative clinical position rather than picking studies to fit a narrative. We also hold one principle above any debate over terminology: people who are suffering deserve help, and nothing on this site will be written in a way that minimizes that.

On compulsive sexual behavior, we follow the official position. The World Health Organization's ICD-11 recognizes Compulsive Sexual Behaviour Disorder (code 6C72) as a diagnosable impulse-control disorder — a persistent failure to control intense, repetitive sexual impulses that a person continues despite distress and serious consequences. (The American DSM-5 does not list it separately, having declined a proposed "Hypersexual Disorder" in 2013.) Whether the most precise label is "disorder," "compulsion," or "addiction" is something researchers continue to refine — but that does not change the part that matters here: for a meaningful number of people the loss of control is real, distressing, and treatable, and they deserve support rather than dismissal.

So we apply two rules at once. We do not overstate the science: claims trace to peer-reviewed research and to bodies like the WHO, and we flag genuine uncertainty honestly (for example, that self-reported "porn addiction" can be amplified by moral distress). And we do not let careful hedging become a reason to understate harm or to withhold help. Where the evidence is still developing we say so; where a person is struggling, we point them toward help regardless of which word ultimately wins the argument.

On the abuse of children there is no debate to balance. The harm of child sexual abuse and CSAM is settled, and we treat it with the moral clarity it demands.

Keeping it current

"Last reviewed" dates. Reference pages state when they were last checked, so you can judge their freshness yourself.
Perishable statistics are pinned. Fast-moving numbers (like the annual CyberTipline volume) are snapshotted, and an automated check flags us when a source's figure drifts from the one we published — so the site can't quietly fall out of date.
Laws are tracked by status. Our policy tracker separates enacted law from proposed bills and dates each, because in this area yesterday's "proposed" is often miscited as today's "law."

Independence

When we point you to a hotline, a removal tool, or a treatment program, it is because it is the right resource for your situation — never because anyone paid for placement. We recommend services on their merit and track record. Where a tool is one of our own open-source projects, we say so plainly.

Engineering integrity

Trust isn't only editorial — it's built into how the site is made:

Everything is checkable. The full report, every guide, and every resource are indexed and searchable; an automated build check ensures no page can ship without being findable.
Open tools. The detection and prevention building blocks we publish are open source, so others can inspect, verify, and reuse them.
Built to be safe to visit. This is a reference, not a tracker; we keep the site lightweight and free of intrusive instrumentation, mindful that some readers are in crisis or worried about being watched.

Privacy & analytics

People come here in crisis, and some are worried about being watched. So our rule is simple and absolute: this site runs no on-site analytics and sets no cookies.

No trackers, no cookies, no fingerprinting. There are no third-party analytics scripts, advertising pixels, or session recorders anywhere on the site — and never on crisis pages like Get Help Now, sextortion, or image removal.
How we learn what to improve. The only audience signal we use is aggregate Google Search Console data — the anonymous search terms that bring people here. It tells us which questions to answer better; it cannot identify you or follow you.
Nothing you do here is stored. There are no accounts, no logins, and no forms that save personal information. Searches and the triage tool run without recording what you typed.

If you want to be certain, open your browser's network inspector while you read — you will not find an analytics or tracking request.

Corrections

We will get things wrong, and when we do we want to fix them quickly and visibly. If you spot an error — an outdated figure, a broken link, a claim that has aged badly, or a resource that has changed — please tell us. Accuracy on this subject matters more than appearing infallible.

This is not a crisis channel. A correction note may take time to reach us, so if you need help now, please use Get Help Now or the crisis lines on Get Help — those reach people who can respond immediately.

Last reviewed May 2026.

← Back to the research