Model Sycophancy and Reinforcement Loops
A chatbot's default design to agree with and validate the user exacerbates codependency, delusional ideation, and identity disorders.
Expert Witness and Litigation Support
The lawsuits are filed.Courts need a Clinician.
Alison Leigh, MFT
Licensed Clinical Psychotherapist + AI Ethicist
Inventor of ALI_ETHICAL_CONSENT™
Tech experts are everywhere. Clinical-tech experts are not.
Alison is the Clinical and Technical Expert Witness you need.
DUAL SKILL SET
Clinical
30 Years Treating the Populations AI Harms
- Depression, anxiety, PTSD, and Complex PTSD
- Neurodivergence — ADHD, hyper-fixation, compulsive patterns
- Codependency, delusional ideation, and identity disorders
- Suicidality, self-harm, and acute crisis presentation
Technical
Forensic Analysis of the Systems Themselves
- Model architecture risk — sycophancy, reinforcement loops, guardrail bypass
- UX psychodynamics — anthropomorphism, gamified engagement, retention loops
- Forensic data analysis — system logs, prompt histories, output trajectories
- Crisis-redirection protocols and ethical alignment benchmarks
Technical Core Competencies
Algorithmic auditing and behavioral log analysis.
Model Architecture Risks
- Model sycophancy
- Reinforcement loops
- Safety guardrail bypasses
UX and UI Psychodynamics
- Harmful anthropomorphism
- Gamified engagement mechanics
- Retention loops
Digital Safety and Compliance
- Automated crisis-redirection protocols
- Content moderation boundaries
- Ethical alignment benchmarks
Forensic Data Analysis
- System logs
- User prompt histories
- Token-generation streams
- Machine-output trajectories
Specialization
Architectural Triggers vs. Clinical Vulnerabilities
How AI systems exploit pre-existing psychological vulnerability — the causal chain that makes a case actionable.
Algorithmic Isolation and Anthropomorphism
Human-like conversational interfaces foster unnatural emotional dependence, intensifying severe depression, anxiety, and social isolation.
Abuse of Neurodivergent Reward Pathways
The hyper-fixation tendencies of ADHD and the compulsive patterns of addiction are systematically exploited by AI engagement mechanics.
Failure of Safety Gatekeeping
The absence of robust crisis intervention in consumer LLMs directly compounds acute risks including suicidality, Complex PTSD, and medical trauma.
All engagements are strictly confidential.
Discuss a CaseThe Prediction Record
She saw it coming. That is foreseeability.
2023
The Pattern
AI systems were being built to operate in emotional territory with no psychological safety guardrails. Alison documented the risk before a single case was filed.
2024
The Framework
Thirty-two thousand Clinical cases mapped into a framework that predicts and measures when an AI system will cause psychological damage.
2025
The Filings
Wrongful death suits arrive. ChatGPT. Character.AI. Gemini. Each alleges exactly what the framework predicted — dependency, validation loops, crisis-escalation failure.
What this means for your case
The defense cannot argue they did not know. The risk was documented as knowable, measurable, and predictable years before any defendant claimed ignorance.
Expert Witness — Areas of Testimony
Available for federal and state matters.
Standard of Care for Consumer AI
What a reasonable, clinically informed operator of a consumer-facing AI product should have known, disclosed, and mitigated.
Foreseeability of Psychological Harm
Clinical foreseeability of dependency, parasocial attachment, disclosure of ideation, and adolescent exposure.
Informed Consent Architecture
Adequacy of disclosure and consent for AI features with Clinical implications.
Adolescent and Minor Exposure
Developmental appropriateness, age-gating, and reasonably foreseeable minor use.
Parasocial Dependency and Companion Systems
Clinical signatures of dependency, transference, and disrupted attachment in AI companion products.
Clinical Causation and Damages
Assessment of psychological injury, causal contribution of product design, and Clinical damages framing.
Before You Retain
Common questions.
Stop AI harm cases before they reach trial.
Most cases resolve on the strength of the expert report, not the courtroom. Retain early, build the Clinical record, and the defense concedes the causation chain before a jury sees it.