Responsible AI for victim services starts with survivor safety

During recent years, AI has dramatically reshaped our digital worlds, and it holds tremendous promise for victim service providers. But it's complicated by a reality every advocate already knows: Victim service providers handle some of the most sensitive information imaginable. Whether you run a domestic violence shelter, a rape crisis center, a sexual assault program, or a family justice center, a leaked intake form isn't an inconvenience. It can be a safety crisis for a survivor.

That's why StriveDB has moved deliberately. We held off releasing anything until we could answer one question with confidence: does this make survivors safer, or does it put them at risk?

Our first AI feature, Import Plus, is live, and more is on the way. The rest of this post explains how we think about responsible AI, how Import Plus actually works, and how the design aligns with NNEDV's guidance and federal confidentiality law.

"AI is going to change every part of how victim service providers operate. The question isn't whether we adopt these tools. It's whether we adopt them in a way that protects the people we serve. Every decision we make about AI at StriveDB starts with the same question: does this make survivors safer, or does it put them at risk? If we can't answer that with confidence, we don't build it."

Daniel Cannon, Founder, StriveDB

How we designed Import Plus for victim service providers

Many of the centers we work with still operate partly on paper. Some are entirely on paper. Some have years of historical files in cabinets that no one has time to digitize. Others use paper intake forms during the actual conversation with a survivor, because they've learned the same thing: a screen between an advocate and someone in crisis adds friction. Eye contact and presence matter more in that moment than data accuracy. The screen can wait.

Import Plus was built around this reality.

An advocate's experience looks like this. You sit with a survivor through a SANE exam, a hospital intake, a crisis call, or an initial protective order meeting. You take notes by hand, or the survivor fills out the form on their own. Later, you scan the form with your phone or a desktop scanner. Import Plus reads it and creates a pre-filled record. Every field is linked to its source location in the document, so you can verify the AI's work at a glance. You check what's there, edit anything that's off, and approve. The conversation was the conversation. The data entry catches up after.

For historical migration, centers can bulk-scan years of paper records. Intake forms, crisis line logs, SANE accompaniment records, hospital advocacy follow-ups, court documentation. Files that would have taken months of manual re-keying enter the database in days.

The human review queue is the part we want to be loudest about. AI never commits a record to your database on its own. A person sees every extraction, can edit any field, and can reject any record before it becomes part of a case file. We considered making auto-commit available as a setting. We decided against it. The risk of an AI mistake propagating into a survivor's record outweighs the time saved.

The review work itself is straightforward. Most centers assign reviewers based on who already handles the data: the advocate who scanned the form, an administrator overseeing client records, or a program manager working through historical batches. A typical page takes under a minute when the form is standard and the AI got most of it right. Forms with unusual layouts, faded paper, or significant handwritten narrative take longer, and some won't extract cleanly enough to be worth reviewing. We'll tell you which forms work well during onboarding.

Import Plus is $100 per month and includes 100 pages of processing. New StriveDB customers receive 1,000 free pages for their initial migration from paper or another system. Additional pages are $0.25 each. It's available on Team, Premium, and Multi-Site plans. If you're on Basic and want to explore AI features, talk to us about whether an upgrade makes sense for your center. Most centers find the math works in their favor once you account for staff time saved on manual data entry. We can help you estimate the savings against your own paper volume before you commit.

How Import Plus answers NNEDV's five questions for AI vendors

We aren't doing this in a vacuum. In late 2025, the National Network to End Domestic Violence published the field's first comprehensive guidance for victim service providers on AI, setting the standard for responsible adoption. The Parasol Cooperative built Ruth, a trauma-informed AI chatbot for survivors based on the same principles used to train crisis advocates. Other vendors in the broader nonprofit space have begun adding AI to their products too, mostly focused on fundraising automation or natural-language reporting. Our work sits inside the case management workflow advocates use every day.

NNEDV's guide includes a checklist of five questions, with this warning: "If the vendor cannot or will not answer these questions clearly and in writing, the tool is likely not appropriate for survivor-facing work." Here's where Import Plus lands on each.

Is user input used to train or improve AI models?

No. We use Amazon Bedrock as our AI infrastructure. AWS contractually guarantees that customer data sent to Bedrock is not used to train foundation models. The model providers we work with see none of your data. Nothing about a survivor's intake form is used to train any AI model.

Where is data stored, and can it be deleted?

Documents and extracted data sit on servers we control, encrypted at rest with AES-256, governed by your StriveDB role-based permissions. Bedrock retains nothing after processing. Records can be deleted on request, following your retention policy.

Can we turn AI features off?

Yes. AI features are sold as opt-in modules. Organizations that prefer not to use AI never have data processed by AI. Within each AI workflow, every record goes through a mandatory human review checkpoint before it becomes part of a case file.

Is the tool VAWA and HIPAA compliant?

StriveDB is HIPAA-aligned. Amazon Bedrock is HIPAA-eligible under AWS's Business Associate Agreement (BAA). We have a signed BAA with AWS and execute BAAs with customer organizations. Because no third party receives identifiable survivor information, the architecture does not trigger VAWA, VOCA, or FVPSA disclosure conditions. StriveDB's security and compliance posture was independently audited by Accorian, which returned no medium or higher vulnerabilities in comprehensive security testing.

Does the AI remember past inputs?

No. Each document is processed as an isolated request. The model has no awareness of any other document, from your organization or any other.

NNEDV's framework is the floor, not the ceiling. NNEDV calls one of these commitments "AI is assistive, not authoritative," and the human review queue you just read about is one expression of that. Beyond it: no survivor's risk is assessed by an algorithm, and no service eligibility is determined by a model. When a feature could create risk for a survivor, we change the feature. This is why StriveDB doesn't send text messages to survivors, even though most platforms offer it. A text on a shared or monitored phone is a risk. We built QR-code-based forms instead. Every design choice in Import Plus came from the same kind of thinking.

We've put our answers in writing here, and in every Business Associate Agreement we sign with a customer.

What's coming next

Import Plus is the first of several AI capabilities we're building. Three more are on the roadmap.

AI-powered search and filtering. Finding the right record at the right moment in any case management system takes too many clicks. We're building intelligent search that understands natural-language queries against your own data, with the same opt-in, no-training, HIPAA-aligned constraints as Import Plus.

AI-generated summaries of cases and assailant information. Advocates regularly need to brief a colleague, prepare for a hearing, or get up to speed on a case they haven't touched in months. We're building summaries that give advocates the context they need in seconds, generated on demand, never retained at the AI layer.

These summaries are tools for advocates, not surveillance of advocates. Your supervisor doesn't get a different view of your work because AI exists. We're also paying close attention to subpoena risk and discovery exposure as we design these features.

A support chatbot for StriveDB users. Software questions shouldn't always have to wait for a support ticket. We're building a chatbot that helps advocates and administrators find answers about StriveDB features, workflows, and reporting, with the same opt-in and privacy controls as our other AI tools.

All three will follow NNEDV's framework. None will be released until it can pass NNEDV's vendor-question audit.

Frequently Asked Questions