A prototype progress update from DigiCivic Initiative, June –July 2026

 

The problem this prototype is trying to solve

 

When a woman in Nigeria is targeted by a deepfake, or has intimate images shared without her consent, the abuse itself is often not the slowest part of her ordeal. The slow part is what comes after: collecting screenshots before they disappear, working out who posted what and when, turning a traumatic personal account into something a court will recognise as evidence, and waiting, sometimes for weeks, for a lawyer to prepare a case.

 

This kind of abuse is known as technology-facilitated gender-based violence, or TFGBV, harm that uses digital tools (social media, messaging apps, fake accounts, AI-generated images) to threaten, harass, or expose someone, usually a woman or girl. It’s a form of what’s more broadly called online gender-based violence (OGBV).

 

Digital harm moves at the speed of a repost. Legal response, in most places, does not. That mismatch is the problem Àdàbà AI is being built to close: not by replacing a lawyer’s judgment, but by speeding up the repetitive, time-consuming groundwork; organising evidence, working out what kind of legal case it might support, and preparing a first draft of the paperwork, that currently stands between a survivor’s report and a lawyer’s desk.

 

What “prototype” means at this stage

 

Àdàbà AI is being developed by DigiCivic Initiative as part of the A+ Alliance Gender & AI Innovation Collective, a group of feminist AI projects supported by Code for Africa. June and July were spent on discovery and design: understanding the problem alongside the people who will actually use the tool: survivors, lawyers, and women’s rights organisations (WROs), before any part of the system gets built.

 

June was foundational. The team finalised a five-month work plan (understand the problem → design the system → build a test version → test it → run a small real-world pilot) and produced the first documented user journey, reviewed with legal, technical, and programme staff.

 

July moved from plan to structure. Two things happened that shaped the tool more than anything else so far.

 

Key design decision: separating “how ready is this evidence” from “how urgent is this survivor”

 

The most important decision to come out of July’s consultations was a refusal to answer two different questions with one number.

 

A case can arrive with almost no evidence and still be an emergency; a survivor reporting that a partner has threatened her life after sharing her images has little that can be neatly documented, but cannot wait. A case can also arrive with excellent, well-preserved evidence and be low-risk. For example, an old case of impersonation where the abuse has already stopped.

 

If those two things were combined into a single score, a well-documented but low-urgency case could end up looking more important than an urgent one with less evidence, or the reverse. So the framework that came directly out of July’s consultation with lawyers keeps them separate:

 

• An Evidence Completeness Score – how much of what a lawyer needs is actually there: the survivor’s account, who’s involved, the digital evidence itself, whether that evidence has been preserved intact, its relevance to the law, and any supporting material like witness statements or police reports.
• A Survivor Risk Assessment – how urgently the survivor herself needs help, based on things like immediate physical danger, how severe the abuse is, whether it’s getting worse, particular vulnerabilities, psychological impact, and how widely the abuse has been shared.

 

Each produces its own result. Neither is allowed to stand in for the other. A lawyer opening a case sees both scores, plus a recommended next step rather than one flattened number that hides which problem is actually the most urgent.

 

This is a decision worth borrowing directly for anyone building a similar tool: how complete the evidence is, and how urgently a person needs help, are two different questions. Squeezing them into one score will always hide one or the other.

 

Community feedback, built in rather than bolted on

 

July’s consultations happened in stages, not as a single approval meeting. A session with women’s rights organisation partners on July 2 was followed by a lawyers’ consultation on July 24 – the same session that produced the scoring framework described above. That framework wasn’t built by the technical team and then shown to lawyers for a stamp of approval; it was built together with them, in the room.

 

The design sketches (wireframes) for the tool followed the same approach. Review sessions with stakeholders in July led to concrete changes the team is now working through, with further reviews already scheduled rather than treated as a one-time sign-off.

 

Not everything went to plan. A session meant to bring survivors directly into the design process didn’t go ahead in July – it has been postponed for safety reasons due to recurring attacks they faced and became detached. Designing a tool with trauma survivors in mind means accepting that their safety comes before the project timeline.

 

Where the technical design is heading

 

Two decisions carried over from this period’s design work matter most, because they’re what make this project reusable by others rather than just useful to DigiCivic:

 

• A human always makes the final call. The tool’s process – a survivor or case worker starts a report, evidence is uploaded and digitally “sealed” (using a technique called hashing, which proves a file hasn’t been altered since it was uploaded), the AI reviews it and suggests what kind of case it might be, and drafts a first version of the legal paperwork, is built so that AI output is clearly labelled “draft only, needs lawyer review” the moment it’s produced. This isn’t left to staff to remember to check; it’s built into the system itself.
• A “plug-and-play” design. Rather than locking the tool to one specific AI company or one specific storage company, the system is built with swappable parts: whichever AI service does the analysis, whichever service stores the files, whichever service sends notifications, can each be changed independently. One organisation might run it on a paid AI service and cloud storage; another, with a smaller budget, could run the same core system on a free, self-hosted alternative. The actual case-handling logic, how a case moves from report to draft affidavit, stays the same either way.

 

That second point is the practical takeaway for any organisation considering something similar: build the core system so it doesn’t depend on one specific paid tool. That’s the difference between something only one well-funded organisation can run, and something a smaller women’s rights organisation could realistically adopt.

 

Why this matters beyond one organisation

 

Before building anything, DigiCivic Initiative looked at what already existed in Nigeria’s digital rights space. Platforms like TechHer’s Kuram and Paradigm Initiative’s Ripoti already give survivors safe ways to report OGBV and other digital rights violations. Àdàbà AI is deliberately designed to start after that first report is made, it isn’t another place to report abuse, it’s the next step: turning an already-reported case into a case file a lawyer can actually use in court. The aim is for these tools to work together, each handling a different part of a survivor’s path to justice, rather than competing for the same person’s attention.

 

Looking ahead

 

August and September are about moving from design to a working system: finalising the technical architecture, turning the wireframes into an actual working prototype, and testing the evidence-handling and triage features against the scoring framework described above. The next progress post will cover what held up under testing, how the Evidence Completeness and Risk Assessment framework performed against real (simulated) cases, and what needed to change. The prototype and its documentation are on track to be published as open source in September 2026, along with a guide for other organisations adapting it to their own legal and language context.

 

Àdàbà AI is developed by DigiCivic Initiative as part of the A+ Alliance Gender & AI Innovation Collective, supported by Code for Africa. This post is published under a Creative Commons Attribution 4.0 International licence (CC BY 4.0) – reuse and adaptation are encouraged with attribution.