AI-Powered Prevention of Non-Consensual Intimate Image Sharing

A case study on StopNCII.org: how hash technology (digital fingerprinting) and cross-platform collaboration are redefining digital safety for survivors of image-based abuse.

Context & Problem Statement

Non-Consensual Intimate Image (NCII) sharing — commonly known as revenge porn — is a form of image-based sexual abuse where intimate images or videos are shared without the consent of the person depicted. This includes threats to share such content, which create ongoing psychological terror and coercive control. The problem has grown exponentially with the rise of smartphones, social media, and encrypted messaging platforms.

300,000+

individual NCII images removed by Revenge Porn Helpline

90%+

successful removal rate achieved

2015

Revenge Porn Helpline established (UK)

Traditional legal remedies are often slow, cross-jurisdictional challenges are immense, and survivors face re-traumatisation by having to repeatedly report their images to different platforms. As Nicola Henry from RMIT University notes: “Victims have three key justice needs: recognition of harm, an array of options to choose from, and most importantly, the ability to reclaim control of their lives. A key priority is having their content removed or taken down quickly.” StopNCII.org addresses this gap through AI-powered hash technology.

Intervention Overview: StopNCII.org

StopNCII.org is a free tool designed to support victims of Non-Consensual Intimate Image (NCII) abuse. It is operated by the Revenge Porn Helpline (RPH), which is part of SWGfL (South West Grid for Learning), a not-for-profit charity founded in 2000 with a mission to ensure everyone benefits from technology, free from harm.

AI Technology:

Image hashing (digital fingerprinting) creates a unique hash value for each image/video. Duplicate copies produce identical hashes, enabling detection without storing the original media.

Privacy-First Design:

Hashes are generated directly on the user’s device. StopNCII.org never downloads or stores the original images. Only the hash (digital fingerprint) is shared with participating companies.

Cross-Platform Collaboration:

Hashes are shared with participating tech companies (social media platforms) so they can proactively detect and remove matching images.

Eligibility criteria: The tool is available to individuals who are (1) the person in the image, (2) 18 years or older at the time the image was taken, (3) currently over 18, (4) still in possession of the image, and (5) depicted nude, semi-nude, or engaging in a sexual act.

AI Technology Explained: Image Hashing as a Digital Fingerprint

Technical insight for M&E professionals: Image hashing is a deterministic algorithm that converts an image into a fixed-length string of characters (the hash). Even if the image is resized, recompressed, or slightly edited, perceptual hashing can still recognise near-duplicates. This is distinct from cryptographic hashing (e.g., SHA-256) where any change produces a completely different hash. StopNCII.org uses perceptual hashing appropriate for content moderation at scale. The hash serves as a ‘blocklist’ that platforms can query without ever seeing the original content — a privacy-preserving application of AI.

Step 1 (User device): User selects the intimate image(s)/video(s) on their own device. The StopNCII.org tool generates a unique hash (digital fingerprint) locally. The original media never leaves the user’s device.
Step 2 (Submission): Only the hash value is transmitted to StopNCII.org. No metadata, no image data, no personal identifiers beyond what is minimally required.
Step 3 (Distribution): StopNCII.org shares the hash with participating companies (social media platforms, hosting services) that have committed to using the hash database for proactive detection and removal.
Step 4 (Platform action): Participating companies run the hash against content uploaded to their platforms. When a match is found, the content is flagged and removed according to their terms of service and local laws.
Step 5 (Outcome): The user’s intimate images are prevented from spreading, and future uploads of the same content are blocked. The user receives confirmation but never needs to re-upload or re-share their images with any authority.

Theory of Change

Inputs	Activities	Outputs	Outcomes	Impact
Hash algorithm (AI) Participating platforms Helpline staff (RPH) Privacy-by-design infrastructure Funding from donors/partners	Generate hashes locally Submit to StopNCII.org Share hash database with platforms Platforms run detection Remove matched content	Hash database created Images removed from platforms User receives confirmation Prevention of re-upload	Survivors regain control Reduced re-traumatisation Deterrence of perpetrators Cross-platform consistency	Reduced NCII harm Stronger digital safety ecosystem Progress toward SDG 5 (Gender Equality) and SDG 16 (Peace & Justice)

Evaluation Objectives

Measure the technical effectiveness of perceptual hashing in detecting NCII images across different platforms and image variations.
Assess user trust in the privacy-preserving design (no upload of original images).
Evaluate the speed and consistency of removal across participating companies.
Analyse the accessibility of the tool across different demographics, languages, and digital literacy levels.
Examine the psychological outcomes for survivors (sense of control, reduction in anxiety).
Identify gaps in platform participation and develop recommendations for expanding the hash database network.

Key Evaluation Questions

Effectiveness:
Does the hash-based detection achieve >90% removal rate across all participating platforms?

Relevance:
Does the eligibility criteria (18+, own the image) exclude vulnerable groups such as minors or those whose images have already been shared widely?

Privacy & Ethics:
Is the local hash generation truly secure? Could hashes be reversed or used to identify users?

Equity:
Are there language barriers, digital access barriers, or regional limitations (e.g., certain countries blocked)?

Evaluation Methodology (Proposed)

Quantitative analytics: Number of hashes submitted, number of removals per platform, time-to-removal metrics, false positive/negative rates.
User experience surveys: Embedded feedback forms (optional, anonymised) to capture survivor satisfaction, perceived control, and trust.
Platform compliance audits: Test submissions to measure whether all participating companies consistently action hashes within agreed SLAs.
Comparative analysis: Compare outcomes for users of StopNCII.org versus traditional reporting mechanisms (legal, platform-specific reporting).
Qualitative interviews: With a subset of users (opt-in, with trauma-informed protocols) to understand psychological impact and areas for improvement.
Stakeholder interviews: With platform representatives, helpline staff, and endorsing organisations (e.g., National Center for Victims of Crime, Take Back the Tech).

Key Findings (Based on RPH Operational Data)

High removal effectiveness: Over 300,000 individual NCII images removed, with a removal rate exceeding 90%.
Privacy-first design builds trust: Users are more willing to engage when they retain control of their images and do not have to upload them to a third party.
Cross-platform collaboration multiplies impact: A single hash submission can protect the user across multiple social media platforms simultaneously, reducing the burden on survivors.
Proactive prevention is more effective than reactive takedown: Once a hash is in the database, platforms can block re-uploads, preventing cycles of abuse.
Global endorsements validate the model: Organisations from Sri Lanka, South Korea, Brazil, Nigeria, Denmark, Hong Kong, Tanzania, Iceland, Australia, and the USA have publicly supported StopNCII.org, indicating broad applicability across legal and cultural contexts.

Challenges & Limitations

Eligibility exclusions: Individuals under 18 cannot use the tool, and those whose images have already been widely shared without their possession of a copy cannot generate a hash.
Platform participation is voluntary: Not all tech companies have signed up, creating gaps where NCII content may still spread.
Hash evasion techniques: Sophisticated perpetrators may alter images sufficiently to evade perceptual hashing (e.g., adding noise, cropping, recolouring).
Limited to image/video matching: The tool does not address threats, harassment, or other forms of image-based abuse that do not involve matching existing media.
Geographic and language accessibility: While the tool is global, language barriers and varying levels of digital literacy may limit uptake in some regions.

Lessons Learned for AI-Powered Social Good

Privacy by design is non-negotiable in sensitive contexts. StopNCII.org succeeds because users retain full control of their intimate images. Any AI system handling sensitive data must minimise data collection and maximise local processing.
Technology alone is insufficient without survivor-centred design. The tool was built on decades of helpline experience (RPH since 2015) understanding survivor needs, fears, and trauma.
Cross-sector collaboration multiplies impact. A single NGO cannot force platforms to act; but a shared hash database creates collective responsibility and standardised protection.
AI for social good requires ongoing evaluation of harms and exclusion. Eligibility criteria (e.g., age restrictions) may protect some but exclude others. Continuous equity audits are essential.
Measuring success goes beyond removal counts. Psychological outcomes, sense of agency, trust in digital systems, and deterrence effects are equally important indicators.

Recommendations for Scaling and Improvement

Expand platform participation: Advocate for all major social media, messaging, and cloud storage platforms to join the hash database network.
Develop pathways for minors: Create a parallel trusted intermediary process for under-18s, involving parents, guardians, or child protection agencies.
Investigate perceptual hash robustness: Fund research into adversarial evasion techniques and develop next-generation hashing algorithms resilient to manipulation.
Enhance multilingual and low-literacy access: Translate the tool into more languages (currently available in English and Traditional Chinese) and provide video/audio guidance.
Build integration with legal pathways: Offer users optional, anonymised data to support law enforcement without compromising their privacy or willingness to use the tool.
Commission independent evaluation studies: Publish third-party evaluations of effectiveness, equity, and user outcomes to build trust and guide iteration.

Implications for M&E Practice in AI & Digital Safety

Measuring AI effectiveness:
Evaluators must assess both true positives (actual NCII removed) and false positives (legitimate content flagged).

Ethical frameworks are critical:
AI systems handling sensitive content require evaluation of privacy preservation, informed consent, and user agency.

Real-time monitoring:
Dashboard metrics (hashes submitted, removals, response times) should be part of a live M&E system, not just periodic reports.

Survivor-centred indicators:
Beyond technical KPIs, evaluate user trust, psychological relief, and sense of control.

Key Takeaways for EvalCommunity Members

AI (perceptual hashing) can be deployed in privacy-preserving ways that protect, not exploit, vulnerable users.
Cross-platform collaboration multiplies the impact of AI moderation tools; evaluation must assess network effects and participation gaps.
Technology alone is insufficient: helpline support, survivor co-design, and trauma-informed pathways are essential complements.
Eligibility criteria (e.g., age, possession of images) can create exclusion. Evaluators must identify and advocate for those left behind.
StopNCII.org demonstrates that AI for social good is possible when privacy, transparency, and survivor agency are treated as core product features, not afterthoughts.

Disclaimer: This case study presents an independent analysis of StopNCII.org, operated by the Revenge Porn Helpline (RPH) / SWGfL. It is based on publicly available information from the StopNCII.org website and RPH documentation. This analysis reflects the perspectives of EvalCommunity on strengths, gaps, and recommendations for M&E practice in AI-powered digital safety. EvalCommunity does not claim ownership of the original program or content. Readers are encouraged to consult the original StopNCII.org website and RPH sources for complete context and current operational data.

References & Sources

StopNCII.org. (n.d.). How does StopNCII work? Retrieved from https://stopncii.org
Revenge Porn Helpline (RPH / SWGfL). Operational statistics and documentation.
Henry, N., & Flynn, A. (2020). Image-based sexual abuse: A global framework for redress and prevention.
Girls Not Brides. (2025). Meet Kemi, the WhatsApp chatbot supporting survivors in West Africa.

© EvalCommunity — Case study on StopNCII.org / Revenge Porn Helpline (SWGfL). For professional M&E training on AI ethics, digital safety, and gender-based violence prevention, visit EvalCommunity Academy.

StopNCII