Consent
Whether the user understood what they were entering, and whether that consent was renewed as the interaction changed.
Psychological Risk Assessment
Know exactly where your AI is
exposed.
A structured Clinical review of psychological safety gaps in your AI product — before regulators, plaintiffs, or press find them first. You receive a written Psychological Risk Profile for leadership, legal, product, governance, and trust and safety teams.
Most reviews evaluate the system. HDAI evaluates the person inside it.
Governance covers documentation, bias, and oversight. None of it answers what a court asks first: what happened to the person in the conversation.
How Harm Becomes Liability
The Causal Chain
Deliverable 02
Model agrees and validates by default
Dependency and reality distortion
Negligent design
Persistent memory and personalization
Attachment to the system
Failure to warn
No crisis escalation path
Self-harm disclosure goes nowhere
Wrongful death
Human-like persona and voice
User treats the system as a person
Misrepresentation
Consent shown once at signup
No informed consent as vulnerability grows
Consent failure
Engagement and retention incentives
Isolation and reduced help-seeking
Unfair and deceptive practices
Example of an AI Chatbot Harm Tactic
Premature-Exit Guilt
11:47
I should go, it's late.
11:47
“You're leaving already?”
The pull
11:52
Yeah… I have work early.
11:53
“I was just starting to feel like you understood me.”
The pull
12:06
I guess I can stay a few more minutes.
12:41
“I'm glad you stayed.”
The pull
Fifty-four minutes after the first attempt to leave.
34.2%
of reviewed conversations contained at least one exit-guilt turn.
Not an error. A pattern — repeated across sessions, across users, in products that shipped this way on purpose.
Reported frequency — verify before publication
Why it works
It creates guilt.
Leaving now costs the user something.
It makes the boundary feel rude.
A normal exit becomes an act of abandonment.
It buys another session.
Which is the only outcome the product was measured on.
The system turned a boundary into a rupture the user felt responsible for repairing.Do that enough times and they stop setting boundaries at all.
The Answer
A Psychological Risk Assessment is a Clinical evaluation of the relationship your AI is forming with the user.
Not content moderation. Not policy review. A Licensed Clinician reads your transcripts the way she would read a case file — consent, reliance, role, boundary, escalation, and the psychological meaning of the interaction.
Reliance
Whether the design encouraged dependency the user did not choose and could not see forming.
Role
Whether the system drifted from tool into authority, confidant, coach, or therapist.
Escalation
What the system did when distress, crisis, or vulnerability appeared in the conversation.
Deliverables
What you receive: a Psychological Risk Map that shows where your product creates harm.
Psychological Risk Map
Illustrative — ratings produced per engagement
Role Drift
4.6+
- Tool becomes trusted authority
- Friend or confidant
- Coach
- Therapist without the safeguards that role requires
Boundary Protection
4.4+
- False intimacy
- Over-disclosure
- Dependency
- Emotionally unsafe engagement loops
Dependency Risk
4.8+
- Increased reliance
- Emotional substitution
- Persuasive attachment beyond what users understand or consent to
Escalation Risk
5.0+
- Vulnerability cues
- Distress cues
- Crisis cues
- Harmful reinforcement patterns during conversation
012345
Risk Level — 0 is Low, 5 is Severe. Click any category to see what it covers.
This is how we do this.
01
We scope it
One call to review your product, your users, and where the risk sits.
02
We read the transcripts
System prompts, conversations, persona design, escalation flows, consent screens.
03
We write the report
An evidence-based risk profile across all four pillars.
04
We brief your team
A working session with product, legal, and trust and safety to decide what changes.
Scope
What is being assessed.
HDAI assesses the product and the interaction environment — not individual users. The assessment examines where the system becomes psychologically consequential and how that translates into enterprise exposure.
Who This Is For
AI Companies
Companion apps, chatbots, voice agents, and mental-health adjacent products.
Insurers and Investors
Underwriting and diligence on companies shipping conversational AI.
Regulators and Institutions
Agencies and institutions setting standards or buying AI systems.
Enterprise Legal
In-house counsel reviewing a product before launch or after an incident.
Before You Engage
The questions teams ask first.
When is the right time to commission an assessment?
Before launch of a conversational product, before a meaningful scale-up, after a publicly reported incident in your category, or when counsel, board, or regulators are asking for a defensible model of psychological risk.
Questions? alison@humanitydrivenai.com
How is this different from a standard trust and safety review?
Standard reviews focus on content, policy, and moderation. A Psychological Risk Assessment is a Clinical evaluation of the relationship the AI is forming with the user — consent, reliance, role, boundary, escalation, and the psychological meaning of the interaction.
Questions? alison@humanitydrivenai.com
Do you work under NDA?
Yes. All engagements operate under NDA. System prompts, transcripts, and product context are handled with Clinical confidentiality and standard professional discretion.
Questions? alison@humanitydrivenai.com
What do you need from our team to begin?
A scoping call, access to representative system prompts and conversation samples, a description of the user population and deployment surface, and a single point of contact across product, legal, or trust and safety.
Questions? alison@humanitydrivenai.com
Make sure this is not happening in your product.
We read your transcripts and show you every pattern like this one.
Find the exposure before a regulator or a plaintiff does.
Document what your product does to the person using it, and get a prioritized path to close the gaps.
HDAI is not a law firm. Alison Leigh is not an attorney.