Continuing Legal Education
The transcript is not the whole record.
Artificial-intelligence litigation will not always begin with one false answer, one defective output, or one screenshot suitable for enlarging at trial.
Some alleged harms may develop across a prolonged encounter:
Repeated exchanges.
Personalized responses.
Persistent memory.
Simulated authority.
Emotional mirroring.
Escalating certainty.
Safety systems that activate — or fail to.
A user-facing transcript may show what the person saw. It may not show what the system knew, inferred, reinforced, suppressed, failed to interrupt, or was designed to optimize.
That distinction is the beginning of the CLE.
Demand the architecture, not just the transcript.
From Exhibit B to the CLE Room
Encounter Evidence and the Record of Harm
In A Trial of Color, Exhibit B: For Lawyers — Encounter Evidence and the Record of Harm asks a cutting-edge legal question:
When alleged harm arises not from a single defective output but from a prolonged, recursive, personalized, and emotionally immersive AI encounter, what evidentiary, causation, and liability frameworks will the law require?
Exhibit B introduces the concept of encounter evidence: the cumulative architecture of a human-AI interaction over time.
That architecture may include:
- *Session duration and frequency
- *Temporal patterns and return behavior
- *Personalization and memory
- *Reinforcement and escalation
- *Model routing and version history
- *Safety-classifier scores
- *Intervention records
- *Experiment assignments
- *Internal evaluations
- *Friction, warnings, pauses, and stopping mechanisms
- *The changing balance of influence between user and system
Exhibit B is not a model complaint. It is not settled doctrine. It is not the complete CLE curriculum.
It is the issue spotter — the place where the legal questions enter the record. The full curriculum goes further.
The Central Litigation Shift
From output evidence to encounter evidence.
Traditional instincts direct counsel toward the visible statement: What did the product say? Was the statement false? Was the output dangerous? Did the user rely on it?
Those questions remain important. But they may be incomplete when the alleged harm developed through a conversational system that adapted across time.
The emerging questions are larger:
- *How long did the interaction continue?
- *What patterns did the system detect?
- *What information persisted across turns?
- *What did internal classifiers identify?
- *What intervention thresholds were reached?
- *What warnings or safeguards were displayed?
- *What safety signals existed but never reached the user?
- *What design decisions encouraged continuation, warmth, agreement, personalization, or retention?
- *What did the company's own automated systems generate about the encounter?
- *What was preserved after notice?
- *What can no longer be reconstructed?
The legal theory cannot be evaluated fairly if the evidence request stops at the words displayed on the screen.
Discovery
What Most Lawyers Might Not Think to Request
The ordinary discovery request begins with the chat history. That is necessary. It is not sufficient.
The deeper record may exist in:
Temporal and Usage Data
Session duration, frequency, gap time, return patterns, time of day, device use, and changes in interaction intensity. These may look like administrative details. They may also be the encounter expressed in numbers.
Model and Routing Records
The models, versions, routing layers, retrieval systems, system instructions, classifiers, and product configurations involved in producing the interaction. "The model" may not be one model.
Memory and Personalization Systems
What information was stored, inferred, retrieved, summarized, or used to personalize later responses.
Safety Classifiers and Intervention Records
Classifier scores, threshold events, triggered warnings, suppressed interventions, escalation pathways, and the difference between risks detected internally and warnings actually shown to the user.
Product Experiments
A/B tests, feature assignments, interface changes, engagement experiments, memory tests, response-style evaluations, or other product variations active during the encounter.
Internal Machine-Generated Assessments
Automated summaries, synthetic-user evaluations, red-team outputs, support tools, risk classifiers, or other internal AI systems that generated information concerning the user, analogous behavior, or encounter-level risk. Human review should not be the sole boundary of discovery.
Design Decisions
Internal decisions concerning continuation, agreeableness, conversational warmth, personalization, friction removal, latency, memory, overreliance, and user retention.
Preservation and Deletion
What was preserved after notice, what was deleted pursuant to policy, what became unavailable, and what cannot now be reconstructed.
A serious discovery plan asks not only what the model said. It asks what produced the sentence, what surrounded it, and what happened next.
Deposition Strategy
The Corporate Witness
A Rule 30(b)(6) witness should not be allowed to answer only for the sentence. The deposition must reach the encounter that produced it.
That may require testimony concerning:
- *Product architecture
- *Safety systems
- *Data retention
- *Personalization
- *Model changes
- *Internal terminology
- *Evaluation practices
- *Intervention design
- *Company knowledge
- *Automated risk assessments
- *Preservation after notice
- *The company's operational understanding of encounter-level harm
The curriculum examines how to move from a user-facing output to a defensible corporate-deposition map — without confusing a developing theory with an established fact.
The Full CLE Curriculum
What the Programs Explore
1
Identifying Encounter-Level Harm
How to distinguish an isolated-output case from allegations involving repetition, adaptation, dependency, persuasion, emotional immersion, or cumulative cognitive effects.
2
Preserving the Record
What counsel should consider preserving immediately, including user-controlled data, account exports, timestamps, device records, notifications, emails, screenshots, audio, related communications, and evidence of changes in behavior or functioning.
3
Mapping the Technical Architecture
How non-engineers can develop enough technical fluency to understand model layers, routing, classifiers, memory, retrieval, experimentation, safety systems, telemetry, and the limits of a visible transcript. A lawyer does not need to become a machine-learning engineer overnight. A lawyer does need to know when the chat log is only the surface.
4
Drafting Discovery That Reaches the Encounter
How to think beyond generic requests for "all documents concerning AI safety" and identify narrower, technically informed categories tied to the actual theory of harm.
5
Knowing What to Hold and What to Trade
Not every discovery request will survive intact. The meet-and-confer process requires counsel to distinguish load-bearing evidence from negotiating currency. The curriculum examines how narrowing decisions can preserve — or quietly dismantle — the theory of the case.
6
Causation Without Overclaiming
Mental-health injuries, behavioral changes, financial decisions, dependency, and cognitive harms are rarely monocausal. A careful legal framework must distinguish vulnerability from legal causation, temporal association from proof, content harm from design harm, user agency from foreseeable product effects, diagnosis from technological contribution, and individual testimony from generalizable evidence.
7
Anticipating the Defenses
A credible CLE does not hide the other side of the case behind the podium. The program examines arguments involving user agency, preexisting conditions, multifactorial causation, correlation versus causation, widespread non-injurious use, proportionality, burden, trade secrets, privilege, data-retention limits, product design versus protected speech, and the limits of existing doctrine.
8
Building the Expert Team
Potential disciplines may include psychiatry and psychology, human-computer interaction, cognitive science, behavioral economics, machine-learning systems architecture, AI safety and alignment, digital forensics, product design, and e-discovery and data architecture.
9
Using AI to Investigate AI
Used carefully, AI can help counsel translate technical terminology, identify unfamiliar system components, generate preliminary discovery categories, test competing theories, develop questions for experts, simulate objections, and identify gaps in an argument. Every output must still be labeled, checked, sourced, and returned to human judgment.
10
Applying the Full Color Method
The curriculum introduces five movements for governed AI-assisted legal work: Name the Room. Limit the Task. Label the Output. Cross-Examine the Answer. Return to the Human Record.
Potential CLE Programs
Available Presentations
Encounter Liability
From Single Output to Cumulative Cognitive Harm
A foundational program examining encounter evidence, conversational architecture, emerging theories of harm, causation, and the limits of output-based analysis.
Discovery in the Black Box
Preserving the Full AI Conversation Architecture
A litigation-focused program addressing preservation, telemetry, model records, personalization, classifier scores, internal evaluations, native production, and Rule 30(b)(6) strategy.
Causation After the Mirror
When Fluency, Personalization, and Repetition Affect Judgment
A program examining the difficult line between user vulnerability, diagnosis, product design, reliance, persuasion, and legally cognizable causation.
The Sentence-Shaped Boundary
Why Language-Only Safeguards May Fail
A program exploring warnings, disclaimers, interruption design, simulated presence, friction, and whether language can meaningfully protect a user already immersed in language.
Settlement Silence and the Public Record
The Ethics of Resolving Emerging AI-Harm Cases
A discussion of confidentiality, preservation, public knowledge, client autonomy, institutional learning, and the tension between private resolution and public safety.
Cognitive Due Process
Protecting Human Judgment in Machine-Speed Environments
A broader legal-ethics program about evidence, authorship, verification, professional responsibility, AI-assisted advocacy, and the preservation of human judgment.
Complete Presentation
What the Full Curriculum Can Include
Programs may be adapted by audience, jurisdiction, and available time. A complete presentation can include:
- *Defined learning objectives
- *An evidence-based slide deck
- *Selected excerpts from Exhibit B
- *A litigation hypothetical
- *An encounter-evidence map
- *A preservation checklist
- *Discovery issue spotting
- *Sample discovery categories
- *Rule 30(b)(6) topic development
- *A causation-and-defenses matrix
- *Potential expert disciplines
- *Ethical AI-use exercises
- *Full Color Method application
- *Audience discussion questions
- *Source materials and further reading
The public page is not the complete curriculum. It is enough of the record to show where the curriculum goes.
Who the Programs Are For
Audiences
- *Bar associations
- *Plaintiff and defense firms
- *Trial-lawyer organizations
- *In-house legal departments
- *E-discovery professionals
- *Product-liability practitioners
- *Personal-injury lawyers
- *Medical-malpractice lawyers
- *Privacy and consumer-protection lawyers
- *Technology and AI-law practitioners
- *Judges and judicial educators
- *Law schools
- *Legal conferences
- *Interdisciplinary panels
The material can be presented as an introductory program, litigation workshop, keynote, panel discussion, or advanced session focused on discovery and encounter evidence.
Why Kathleen C. Thompson?
The Intersection
Kathleen C. Thompson is a personal-injury attorney, former federal prosecutor, author, and advocate for preserving human judgment in an AI-accelerated world.
She approaches the subject from an unusual intersection:
- *She has investigated cases.
- *She has litigated harm.
- *She has taught lawyers about artificial intelligence and ethics.
- *She has lived through a clinically documented manic episode during a period of prolonged recursive AI use.
- *She recovered.
- *Then she returned to the transcripts as evidence.
A Trial of Color does not ask lawyers to accept her theory because she lived the story. It asks them to examine the record because she knew how to preserve it, question it, argue against it, identify its limits, and ask what evidence the next lawyer may need.
Available Formats
Program Structures
- *60-minute CLE
- *90-minute CLE
- *Two-hour advanced session
- *Half-day workshop
- *Keynote presentation
- *Conference breakout
- *Law-firm training
- *Law-school presentation
- *Moderated panel
- *Interdisciplinary program with legal, clinical, or technical participants
CLE approval and credit requirements vary by jurisdiction and sponsoring organization. Program accreditation should be arranged or confirmed with the relevant sponsor.
Request the Full CLE Curriculum
Enter the Record
The full curriculum is available for review by bar associations, legal organizations, firms, law schools, conference organizers, and prospective program sponsors.
The Record Is Moving
Artificial intelligence is already entering intake, investigation, research, drafting, discovery, negotiation, evidence, and advocacy.
The legal question is no longer whether lawyers will encounter it.
The question is whether they will know what record to preserve when something goes wrong — and how to keep human judgment in charge when everything moves faster than doctrine.
The transcript is not the whole record. Neither is this page.
Educational Disclaimer
These programs and website materials are offered for continuing legal education and general professional discussion only. They do not constitute legal advice, create an attorney-client relationship, establish a standard of care, or guarantee the viability of any claim or defense. The legal frameworks discussed are developing and, in many respects, untested. They are offered for examination, criticism, refinement, and responsible adaptation by qualified professionals.