The evidence base for CSAM cases.

The Babchishin, Hanson, and VanZuylen (2015) meta-analysis is the most-cited reference for the clinical distinction. CSAM-only offenders are typically young, white, highly educated, and employed in white-collar professions, with significant difficulties in intimate relationships, sexual intimacy, and self-esteem. Mixed offenders (CSAM plus contact) score significantly higher on pedophilic interest than CSAM-only offenders — meaning paraphilic interest, not escalation from legal content, drives the most serious cases (Archives of Sexual Behavior).

Marx, Müller, and Beutel (2021) documented that 5.3% of clinical CSAM users have predominantly adult sexual preferences, suggesting non-pedophilic pathways exist driven by “striving for new and increasingly exciting material” (Cyberpsychology). The takeaway: psychometric profiles are heterogeneous, and a clinically-grounded differentiation can be material to risk assessment, sentencing, and treatment planning.

The Clark et al. (2025) meta-analysis is the current comprehensive reference, covering 21 studies of 15,077 CSAM offenders. Headline numbers over approximately four years: 3.41% for any sexual re-offending, 0.66% for contact sexual re-offending, and 3.05% for CSAM re-offending (Journal of Forensic Psychiatry and Psychology). These are useful in mitigation but should be presented honestly.

The Seto et al. (2011) self-report study is the necessary companion. It revealed that approximately 55% of online offenders admitted to a contact sexual offense at some point in their history, against a 12% rate in official records (Sexual Abuse). Defense and prosecution both have to grapple with the gap between recorded re-offense rates and admitted historical offending. Don't use the Clark numbers alone.

The actuarial tools with the strongest evidence base are Static-99R (static risk factors) and Stable-2007 (dynamic risk factors). Acute-2007 covers proximal risk. These have been validated across multiple populations and are widely accepted in US, UK, Canadian, and Australian courts. They are the foundation of a defensible forensic risk assessment.

Computerized viewing-time tools, particularly the Abel Assessment for Sexual Interest (AASI-3), are widely used but have significant validity controversies. Massachusetts courts ruled the AASI inadmissible in 2002 (upheld 2005). It has struggled to meet the Daubert standard in federal courts. The proprietary scoring algorithm is not transparent, and the foundational Abel & Harlow Child Molestation Prevention Study was never peer-reviewed. Independent studies have concluded the tool is unreliable in adults and lacks sufficient evidence for adolescent use (Wikipedia overview). Defense counsel should be prepared to cross-examine on these grounds when AASI results are part of the prosecution's case.

The Diana Screen (from Abel Screening Inc., Alameda, CA) is marketed primarily to institutions (schools, youth-serving organizations) as a pre-hire or pre-volunteer screening tool. Its evidence base in the peer-reviewed literature is thin and similar caveats apply. The Penile Plethysmograph (PPG) remains the laboratory-grade reference for measuring physiological sexual interest but is invasive, expensive, and used selectively.

See Chapter 03 → Risk factors and forensic profiles for the full instruments discussion.

The Risk-Need-Responsivity (RNR) model is the established framework for matching treatment intensity to assessed risk. Critical: intensive treatment of low-risk offenders may paradoxically increase recidivism by 21% (OJP SOMAPI). Treatment planning premised on “more is better” can be cited in cross-examination as inconsistent with the evidence base.

Specialized programs for CSAM-only offenders include:

• Inform Plus (UK, 25 hours, pre-conviction) — targets pre-conviction populations with documented improvements in pro-offending attitudes, socio-affective functioning, and mental health.
• i-SOTP / i-Horizon (UK, 46–70 hours) — adult sex offender treatment, post-conviction.
• CEM-COPE (Australia, 20-hour group format) — draws from ACT, CBT, and DBT, targets emotional regulation, problematic internet use, and relapse prevention.
• Prevention Project Dunkelfeld / Troubled Desire (Germany / global online) — free, medically confidential. 2024 long-term follow-up showed 0% new CSA among participants without prior CSA history, though an 89.1% CSAM continuation rate (Journal of Prevention). The 0% new-contact-offense figure is significant for mitigation arguments.
• Stop It Now! pilot data confirms benefits in modifying actions to minimize abuse risk (PubMed).

Generic sex-offender programs designed for contact offenders produce little meaningful change in CSAM-only offenders (Australian Institute of Criminology). Routing a CSAM-only client to a generic program rather than a specialized one is a defensible criticism in mitigation.

The TAKE IT DOWN Act (signed May 19, 2025) criminalizes non-consensual intimate imagery including AI deepfakes, with penalties up to three years and a 48-hour takedown requirement for platforms. The ENFORCE Act (passed the Senate unanimously December 16, 2025) equalizes criminal penalties for AI-generated CSAM with traditional CSAM at the federal level (Thorn; Senator Cornyn). The UK Crime and Policing Bill (February 2025) creates an offense for making, adapting, possessing, or supplying a “CSA image-generator.”

First Amendment challenges are live. In U.S. v. Anderegg, a court dismissed possession charges for wholly AI-generated CSAM citing Stanley v. Georgia (private possession), while allowing production and distribution charges to proceed. This is the first federal case heading to appellate court on AI CSAM and the First Amendment (Tech Policy Press). Production and distribution remain charges where the constitutional defense is significantly weaker even before the statutory changes.

In January 2026, Stanford's Center for Internet and Society published a letter to NCMEC analyzing the widely-cited “Generative AI” CyberTipline figures. The headline finding: the frequently-cited 485,000 “AI-related” reports from H1 2025 was misleading. 380,000 of those reports originated from Amazon, and none of Amazon's reports involved AI-generated CSAM. They were hash hits to known CSAM found in AI training datasets. Stanford concluded that nearly 80% of all “Generative AI” CyberTipline reports involved no AI-generated CSAM at all (Stanford CIS).

The practical implication for defense counsel: where a charging document or expert report cites “AI-generated” CyberTipline numbers as scale or aggravating context, the underlying categorization is often unreliable. Demanding the underlying report categorization (and where AI was actually generated vs. matched against existing material) is a reasonable discovery posture. The same caveat applies to news-cycle and policy testimony that relies on aggregate “AI CSAM” figures.

A forensic psychological evaluation by a clinician who specializes in sexual behavior assessment is materially different from a general mental health evaluation, and the difference shows up in court. Look for a doctoral-level clinician with experience in Static-99R / Stable-2007 administration, who treats AASI results with appropriate evidentiary caution, and who can articulate the Risk-Need-Responsivity rationale for whatever treatment recommendation they make.

Bay Area readers: Dr. Jerel Armstrong, Ph.D. (CA PSY27272) maintains a forensic practice specifically for this work, including evaluation and consultation across the full instrument battery (Static-99R, Stable-2007, AASI-3, PPG, MSI-II) and courtroom testimony. Adults and juveniles. Elsewhere, the IITAP CSAT directory locates Certified Sex Addiction Therapists by region for treatment retainer.

Where the case will involve forensic digital evidence, the Project VIC International network has been the de-facto standard for over a decade. Defense can request audit logs and verification reports from any platform-level hash-matching that produced the evidence; see Chapter 06 of the research for the technical landscape.

Mitigation that survives sentencing tends to share a few features: an evidence-based forensic evaluation by an appropriately credentialed specialist, documented engagement with a specialized treatment program (not a generic contact-offender program), a treatment plan calibrated to actuarial risk under the RNR model, and acknowledgment that recidivism risk is low but real.

Arguments that depend on broad claims about “not really hurting anyone” will not survive contact with the evidence base, particularly given the documented ongoing harm to identified victims whose imagery circulates indefinitely and the re-traumatization that occurs each time their imagery is downloaded. Don't run them.

What does survive: clinical heterogeneity (this client is not the worst-case archetype), demonstrated engagement with treatment, the Clark et al. recidivism numbers paired with honest acknowledgment of the Seto self-report gap, and the Dunkelfeld 2024 result on prevention of new contact offenses for the right clinical profile.