AI is an accessibility tool, not a service provider.

What our AI does

• Rewrites public page content (resource descriptions, FAQ entries, our ethics commitments) at a simpler reading level on request, so people with smaller vocabularies, learning differences, or a different first language can understand what the site says.
• Translation pairing for resource search (already shipped in the multilingual search route — EN / ES / TL).

What our AI does not do

• No AI in the DV-survivor flow. DV-survivor and Family Justice Center referrals never call any AI service. Period. Clinical sign-off (Anadora, LCSW) is required before any of that ever changes.
• No AI in the behavioral-health flow. 42 CFR Part 2 information stays inside the human-clinician path. We do not send behavioral-health content to any third-party AI provider.
• No automated decisions about you. AI never decides your trust tier, your credits, your access to a resource, or whether you are approved or denied for anything. Humans do those.
• No personal information leaves our servers. Your name, your address, your phone number, your case notes — none of that is ever sent to a third-party AI provider.
• No crisis advice, no medical advice, no legal advice. The AI rewriter is instructed to refuse anything that looks like a question for those kinds of help and just return the original text. If you are in crisis, call 988 (Suicide and Crisis Lifeline) or 911 right now.

The constraints we hold

1. Every AI feature is opt-in. Nothing AI-powered runs unless you press a button asking for it.
2. Rate limits cap the number of requests an anonymous visitor can make per hour. This is to keep the lights on, not to gatekeep.
3. We do not store the text you ask the AI to rewrite. We do not keep a conversation history.
4. If the AI service is unavailable, we return the original text. You are never left with a broken page.
5. The list of pages the AI is allowed to touch is hard-coded in our source. DV, behavioral-health, recovery, and crisis surfaces are refused at the API layer before any AI provider is called.

What is still open

We have a broader proposal on the table — a multi-population navigation assistant that would help reentry, foster-youth-aging-out, veteran, and recovery users find what they are eligible for. That is materially higher-stakes than what we ship today and is held until (a) clinical sign-off from Anadora, (b) a written safety review, and (c) a real consent surface for any user whose information would flow through it. It is not on by default and not in production.

Questions

See our ethics commitments for the broader non-waivable floor, or our security page for how we handle data more generally.