What responsible AI looks like for CACs

Nearly three in four U.S. adults now use AI in some form. AI isn’t coming to the forensic interview field; It’s already here. A common question our longtime research partner Guardify hears from CAC directors, interviewers, and prosecutors is the right one: how do we use these tools without putting children, cases, or careers at risk?

That question sits at the heart of how Guardify thinks about technology as they work alongside CACs, prosecutors, and law enforcement every day. Their tech-for-good experts see firsthand both the promise and the pitfalls of the AI tools flooding into our field. Below are four AI best practices Guardify offered to share with NCA members like you.

 

---

 

Guardify’s four best practices on smart, safe, practical AI for CACs

 

#1: Know where your data goes

If a tool can’t tell you exactly where your data is stored, who can access it, and how it’s protected, that tool doesn’t belong anywhere near a forensic interview. Before you adopt anything new, ask for the audit reports. Confirm HIPAA and CJIS compliance through the vendor’s Trust Center, not their marketing page. Read the privacy policy itself. (We have seen vendor privacy policy URLs return 404 errors. That is not a small thing.)

A quick rule of thumb on compliance claims: each certification independently takes 3 to 6 months to complete. SOC 2 Type II requires at least six months of documented controls and reviews. CJIS demands a detailed independent audit. That means any organization holding multiple credentials has almost certainly been operating for three or more years. And because each certification carries significant cost, newer companies often simply can’t afford to pursue them. If a vendor is new and claims full compliance, ask for the documentation.

The rule we live by: never upload sensitive case information into public or non-secure AI. ChatGPT, Gemini, Claude, and Perplexity are remarkable tools for general work, but they are not built for child sexual abuse case records. Match the security of the tool to the sensitivity of the data.

When you ask, be specific: request the full audit reports for HIPAA, CJIS, and SOC 2. Most vendors will require an NDA first — that’s normal, because the reports detail exactly how the platform is built. Sign it and read them. For HIPAA specifically, also request a Business Associate Agreement (BAA). If a vendor won’t provide a BAA, they are not a compliant HIPAA partner.

 

#2: Treat AI as support, not replacement

The most useful framing we have heard: treat your AI tools like new team members who happen to have permanent memories. They can be genuinely helpful for digital assistance, administrative work, content drafting, and routine analysis. They cannot replace the judgment, training, and lived experience of a forensic interviewer, advocate, or prosecutor.

Early research is encouraging. Studies are showing that AI can help interviewers generate more alternative hypotheses, ask more productive open-ended questions, and rely less on closed questions. Practitioners report rapid improvement after only a handful of simulated practice interviews with AI-driven avatars. The gains come from using AI as a practice partner and feedback mechanism, not as a stand-in for the work.

 

#3: Match the tool to the use case

Skill development and real case work call for different tools. For training and practice, AI avatars and simulated interview scenarios are showing real promise. For post-interview analysis on actual case data, that work belongs in a platform built for it — with the right access controls, audit trails, and compliance posture baked in.

Guardify’s Evidence Intelligence Suite was designed with this distinction in mind. Transcripts, summaries, question analytics, and disclosure tracking happen inside a HIPAA, CJIS, and SOC 2 compliant environment, with no exposure of case data to public models. The same AI capabilities you might experiment with in a consumer chatbot are available to your team — just inside the secure walls our field requires.

 

#4: Keep Humans in the Loop

Training is not best practice. A workshop, a webinar, or even a certification does not equal mastery. What turns training into skill is the cycle of review and feedback over time, and AI can accelerate that cycle when used well. Use it to spot patterns in your own questioning. Use it to prepare for cross-examination. Use it to flag where protocol drifted. Then bring those observations back into a real human conversation with a peer reviewer or supervisor.

Responsible use sits at the intersection of three perspectives: the tech-savvy team member, the researcher who keeps up with the studies, and the expert practitioner who knows the actual work. None of those three can be skipped.

 

Guardify and CACs in action

Guardify partnered with the Center for Hope at LifeBridge Health, the CAC serving Baltimore, Md., on a webinar called Beyond Avatars: Can AI Enhance Forensic Interviewing Practice and Skill Development?

If you would like the recording, the slides, or just want to talk through how AI could (and shouldn’t) show up in your center, we are happy to share. Our commitment is the same one it has always been: technology that is safe, simple, secure.

 

Interested in learning more?

Visit guardify.com or contact them at to find out how Guardify can support your mission.