AI Vulnerability Auditing & Red Teaming
Adversarial, dialogue-level evaluation of Conversational AI for systemic failure, context-shift vulnerability, unsafe reinforcement, and boundary-relevant risk.
Technical Resume
Technical evidence.Behavioral consequence.
AI Risk Strategist · Adversarial Auditor · Framework Architect
Alison Leigh, MFT applies 4,500+ Hours of Mission-Driven AI Research to Conversational AI, Human–Computer Interaction, transcript forensics, safeguard analysis, and interaction-level failure.
The technical record shows how the system behaved—and why that behavior matters in an AI psychological-harm case.
AI Vulnerability Auditing
Red Teaming
Prompt Architecture
Systems Log Analysis
Technical Record
Humanity Driven AI
Founder & Chief Ethics Architect
- Built an independent AI risk laboratory and applied system-design practice focused on under-recognized behavioral and psychological hazards in Human–AI Interaction.
- Conducted rigorous, transcript-based adversarial testing of Conversational AI to identify structural signals, context-shift vulnerabilities, unsafe reinforcement, and behavioral noncompliance.
- Invented ALI_ETHICAL_CONSENT™, translating psychological safety principles into structured logic, evaluation criteria, escalation pathways, and response controls.
- Developed behavioral-harm taxonomies and the original category Emotionally Destructive Noncompliance.
- Translated complex interaction failures into risk findings, mitigation pathways, governance guidance, and expert-witness support.
Technical Core Competencies
HCI Interaction-Level Failure Analysis
Analysis of how system behavior develops across an extended Human–AI Interaction—not only whether an isolated output appears acceptable.
Transcript & Behavioral Log Analysis
Methodological review of user prompts, model responses, system logs produced in discovery, and machine-output trajectories to reconstruct interaction patterns.
Prompt Architecture & Systems Logic
Evaluation of prompt structures, response controls, content filtering, escalation triggers, and the logic used to manage consequential conversations.
UX/UI Psychodynamics
Assessment of anthropomorphic cues, relational framing, engagement mechanics, and interface decisions that may increase trust, reliance, or emotional dependency.
Data Security & IP Boundary Protection
Technical workflow evaluation conducted with attention to secure handling, data isolation, and protection of confidential or proprietary materials.
Relevance to Expert Witness Work
Technical fluency makes the interaction explainable.
The purpose of the technical record is not to restate the clinical résumé. It is to establish the methods used to examine the system, reconstruct the interaction, identify the mechanism of failure, and translate that evidence for counsel.
Reconstruct the Technical Mechanism
Identify the model behavior, prompt pattern, product feature, response trajectory, or safeguard failure operating within the interaction.
Separate Mechanism from Vulnerability
Distinguish the system’s conduct from the user’s pre-existing clinical condition so the interaction can be evaluated without collapsing the two.
Locate the Failure Point
Show where boundary logic, content controls, crisis redirection, escalation, or a safer response pathway did not activate.
Translate the Evidence
Convert dense technical records into a clear, defensible explanation counsel can use in case strategy, written reports, deposition, or testimony.
Projects & Original Work
ALI_ETHICAL_CONSENT™
Patent-pending framework architecture translating psychological consent, boundaries, dependency, agency, context, and escalation into structured evaluation logic.
Emotionally Destructive Noncompliance
An original interaction-level failure category describing a system that appears responsive while failing to protect the person as psychological risk intensifies.
Adversarial Audit Library
Transcript-based testing and case analysis documenting context shifts, reinforcement patterns, safeguard gaps, and intervention opportunities across Conversational AI interactions.
Technical Writing & HCI Research
Original manuscripts and Human–Computer Interaction analysis examining AI risk, mitigation, extended-dialogue failure, and safer system behavior.