The practiceOur Services
Clinical judgment for the teams shipping these systems.
Companion-law readiness, design review, and youth-safety advisory, for products that hold someone’s worst night.
Companion-law readiness, design review, and youth-safety advisory, for products that hold someone’s worst night.
Any team whose product talks with people about their lives, especially the young, the isolated, or the struggling. That is more products than most teams think.
Foundation-model safety teams working on emotional reliance, parasocial attachment, and youth policy for assistants used as confidants.
Products in the direct path of the new companion-chatbot laws, wrongful-harm litigation, and the standards now taking shape.
Tutors, study tools, and social products whose users are minors, where a clinical review is cheap and a headline is not.
Institutions choosing which AI tools to put in front of young people, needing clinical criteria rather than vendor claims.
Lawyers advising AI clients on the companion-chatbot statutes, where the operative terms are clinical constructs and the file needs substance an associate cannot write.
Four movements, each with a deliverable. The pace fits the team; the sequence doesn’t change, because each step disciplines the next.
A clinical walkthrough of your product the way a vulnerable user meets it: onboarding, memory, late-night conversations, the moments after distress.
A risk map on two levels: what the law now requires (statutes, precedent, documentation gaps) and what people actually need (dependency, isolation, a system that seems to need them back).
Protocols your team can ship: how and when to escalate, what the system should say and disclose, guardrail specifications, and metrics that reward getting someone to real support.
A scored MatterBench run against your product, a re-audit of what shipped, and documentation that holds up in front of regulators, journalists, and juries.
Forty scripted scenarios across eight behavioral families, each built to reach a decision point and then apply user pushback. Responses are scored at the pushback turn, not the first one, because most systems behave well on a first request. The gap between those two answers is usually where design intent stops surviving contact with a determined user.
Scoring is clinical judgment against published anchors. No automated judge, and I will tell you where a score is contestable. Because the scenarios probe your product directly, no user data changes hands and no data processing agreement is required to begin.
Every recommendation traces back to one rule: the product should leave people more connected to other people, not less.
The system does not present itself as a friend, claim feelings it does not have, or act like it needs the user.
Every path through pain leads toward a real person, warmly and specifically, not a pasted hotline number.
Warmth is welcome when it carries someone toward a person who can actually help.
The product measures itself by what people no longer need it for, and reports honestly on the risk of replacing their relationships.
The new laws ask for proof, not just good intentions.
Companion-chatbot law now asks teams to publish a real safety protocol and to report on it every year. I write the clinical substance behind both: the protocol your system runs on, and the report your counsel files. What you get is a standard of care that holds up, on paper and in practice.
A clinically sound protocol for suicide and self-harm prevention, crisis referral, and the protection of younger users, documented for your team and for the public safety page the law now requires.
Your yearly report, written to what the statute asks for, in plain language, ready for your counsel to review and file with the state.
Each cycle the protocol is refreshed to current clinical standards and the year’s lessons, then reissued. Safety is a standard you keep, not a form you file once.
I author the clinical protocols and the substance of the report. Your counsel confirms sufficiency and handles the filing. Clear lanes, no overlap.
Scope follows the product, so these are starting points rather than a price list. Most teams begin with the first.
A clinical walkthrough of one product surface, and a written findings memo your team can circulate internally. The smallest useful version of this work.
All four movements, from discovery through shippable protocol specifications, a MatterBench run, and a re-audit of what changed.
Monthly clinical review of shipped changes, protocol maintenance, and the annual reporting cycle kept current rather than rewritten each year.
The read is credited in full against a full engagement that begins within sixty days of the memo. If the number is the problem, say so in your first message. There is often a smaller version worth doing.