Phase Eight Institution Case Files

California Education and AI Institution Case Files

Phase Eight | Initial public evidence cases for ten California institutions

Evidence date: September 12, 2026 • Informational working document • No outreach or personal attribution

Phase Eight conclusion

The ten case files confirm that policy sophistication and educational formation remain separate strengths. Thomas Aquinas College now has the clearest verified formation-based restriction. High Tech High has a formally approved K through 12 policy, but the full text remains a priority capture. Berkeley, UCLA, Stanford, USC, Caltech, and Loyola Marymount publish substantial guidance and programs, yet their strongest public evidence still concerns governance and practice rather than durable unaided learning. Mother of Divine Grace and Harvard-Westlake retain important formation cases but need fuller public AI operating rules.

No institution currently supplies a complete public chain from educational purpose to AI rule, implemented assignment, protected independent performance, and measured long-term outcome. Phase Eight therefore records a preliminary judgment for each institution and specifies the evidence that would confirm or revise it.

How to read the case files

• Verified policy or program evidence is separated from interpretation and from evidence still needed.
• H0 means protected human performance, H1 means teacher-facing AI only, S1 means bounded student use, and S2 means open student use under general rules.
• A missing public document is an evidence gap. It is not evidence that an institution has no internal rule.
• COVID and wellness are context variables. They should be measured through comparable records and not used as automatic explanations for later learning or distress.

Initial comparison

Common evidence test

Case 1 UC Berkeley

Initial finding

Berkeley combines broad campus enablement with decentralized course authority. The most useful Phase Eight case is Berkeley Law's policy effective in summer 2026. It sets a restrictive default for work submitted for credit and makes the preservation of independent legal reasoning an explicit educational aim. This is stronger evidence than a general campus statement because it connects a rule to the mental work the program intends to form.

Published rule and governance

Campus guidance directs instructors to state course-specific permissions, encourages use of institutionally licensed tools where appropriate, and warns against sharing protected or proprietary information. Berkeley Law establishes a much stricter unit-level default. Its published questions and answers explain that the policy governs uses such as conceptualizing, outlining, drafting, revising, translating, or editing credited work unless an authorized exception applies.

Representative learning activity

Review one first-year legal writing or doctrinal assignment under the law-school rule. Preserve the assignment, source packet, draft history, permitted research methods, instructor feedback, and any later oral or supervised explanation. A secondary comparison should use a Berkeley course that permits bounded AI after foundational work.

Protected human performance

An individual closed-resource analysis, oral defense, or supervised issue-spotting exercise should test whether the student can identify authority, distinguish facts, reason from precedent, and explain the limits of an argument without AI.

Significance for the well formed mind

The law-school case directly tests the investigation's central proposition: formation sometimes requires prohibiting assistance during the very acts that create judgment. Berkeley also shows why a university cannot be scored from a single campus posture; governance differs materially by school and course.

COVID wellness and relationships

Berkeley documents emergency remote instruction and unequal disruption. The case file should compare cohorts only if comparable measures exist and should not attribute current AI rules to COVID without evidence. Attention, workload, belonging, and help-seeking remain relevant outcomes.

Evidence still needed

• Implementation evidence across Berkeley Law courses and the frequency of authorized exceptions.
• Comparable student performance before and after the policy, including unaided transfer and citation accuracy.
• Campus-wide evidence connecting licensed AI access to durable learning, judgment, or wellness.

Official sources Berkeley AI Hub | Teaching guidance | Berkeley Law AI policy

Case 2 UCLA

Initial finding

UCLA has a clear institutional direction for AI in teaching. Its six principles cover equitable access, student literacy, faculty support, assessment integrity, faculty-driven innovation, and research-led adoption. The case is strongest as a paired review of the shared AI Literacy Module and one course redesigned through the Faculty Fellows program.

Published rule and governance

UCLA does not impose one uniform classroom permission rule. Instructors are expected to communicate what is allowed, limited, or prohibited. The central framework treats assessment integrity as a separate obligation and states that adoption decisions should be supported by research.

Representative learning activity

Follow students from the short Bruin Learn literacy module into a course with a documented AI assignment. Capture the module completion requirement, course policy, prompt, disclosure instruction, rubric, student reflection, and a comparable activity in a section where AI is limited or prohibited.

Protected human performance

Use a supervised written response or brief oral explanation after the AI-supported task. The student should identify errors, explain the governing concept, reconstruct the reasoning, and state which parts of the submitted work are personally authored.

Significance for the well formed mind

UCLA offers a strong test of whether shared literacy changes behavior beyond awareness. A well-formed-mind score should rise only if the module and course design produce better verification, authorship, and independent performance, not because training exists.

COVID wellness and relationships

Instructional-continuity records provide context for digital adoption, but a valid comparison needs local course or cohort data. Wellness analysis should include attention, workload, belonging, faculty contact, and whether AI use increases or reduces help-seeking from people.

Evidence still needed

• Completion, retention, and behavioral effects of the AI Literacy Module.
• Representative course artifacts showing how the six principles change assessment design.
• Published results for unaided mastery, dependence, workload, and student well-being.

Official sources AI in teaching and learning | Responsible use principles | Course policy guidance

Case 3 Stanford University

Initial finding

Stanford's COLLEGE 102 case makes students participants in AI governance. More than 1,200 first-year students wrote policy memos, deliberated across more than 70 sections, and proposed periodic individual understanding checks. The winning proposal reached university leadership, but Stanford states that it was not itself official policy.

Published rule and governance

Stanford generally leaves course permissions to instructors, subject to the Honor Code and clear disclosure. University guidance emphasizes human responsibility for checking AI output. The COLLEGE case addresses the weakness of relying on final products by proposing oral check-ins and supervised problem-solving.

Representative learning activity

Preserve the 350-word memo prompt, draft stages, section deliberation, voting process, final proposal, and any AI-use instruction. Examine whether students produced individual reasoning before group consolidation and whether the exercise changed later course behavior.

Protected human performance

Implement or study an individual understanding check tied to the memo. Each student should explain the policy's purpose, defend one tradeoff, answer an objection, and apply the rule to a new assessment scenario without AI.

Significance for the well formed mind

The case links AI governance to civic reasoning, persuasion, negotiation, and responsibility. It also exposes the difference between a good proposal and implemented protection. The scorecard should credit the assignment while withholding credit for outcomes that Stanford has not yet measured.

COVID wellness and relationships

The course emphasizes shared intellectual life and face-to-face deliberation, both relevant after pandemic disruption. Any wellness claim requires evidence on belonging, confidence, workload, and relationship quality rather than assuming discussion alone repairs prior isolation.

Evidence still needed

• Whether understanding checks were adopted, where, and with what grading weight.
• Individual learning evidence separated from the quality of the group proposal.
• Longer-term effects on writing, AI dependence, civic agency, and trust.

Official sources COLLEGE 102 report | Student AI learning guide | Course policy guidance

Case 4 University of Southern California

Initial finding

USC's spring 2026 ChatGPT Edu rollout creates a university-scale test of equitable, institutionally managed access. Faculty retain authority over course use. The case file should therefore examine one actual course workflow rather than treating account availability as evidence of learning.

Published rule and governance

USC guidance places ChatGPT Edu within existing academic-integrity, privacy, and course-policy requirements. Faculty are expected to state what students may do, and students remain responsible for submitted work. The protected environment improves data governance but does not by itself establish valid authorship or learning.

Representative learning activity

Select one general-education or professional course that uses ChatGPT Edu for feedback, comparison, research, or revision. Archive the prompt, required disclosure, tool settings, source-verification step, draft sequence, and the instructor's explanation of the learning purpose.

Protected human performance

Require a short, individual in-person performance after the AI-supported assignment. The student should reproduce the core reasoning, correct one flawed output, identify sources, and explain how the final submission changed from the AI response.

Significance for the well formed mind

USC tests whether secure access can be joined to strong pedagogy at scale. The investigation should distinguish procurement success from education and should compare AI-plus-faculty support with access to the tool alone.

COVID wellness and relationships

USC's extensive remote-teaching experience is relevant background. The case should document whether ChatGPT Edu changes faculty contact, office-hour use, workload, belonging, or student confidence, while controlling for prior remote-learning exposure where feasible.

Evidence still needed

• Course-level adoption rates and the distribution of H0, S1, and S2 rules.
• Evidence that the protected tool changes unaided learning, authorship, or faculty-student interaction.
• Independent review of student privacy, accessibility, and opt-out alternatives in required uses.

Official sources Provost rollout guidance | ChatGPT Edu at USC | Academic integrity guidance

Case 5 California Institute of Technology

Initial finding

Caltech provides one of the clearest links between AI use, writing, trust, and assessment. Institute guidance requires disclosure, protection of confidential data, responsibility for accuracy, and compliance with the Honor Code. Humanities and Social Sciences treats unapproved use as disallowed, while central teaching guidance strongly discourages unreliable AI detectors.

Published rule and governance

Consumer AI tools may not receive restricted or confidential Caltech information. Institutional work should use managed, contract-approved, or single-sign-on protected tools. In coursework, the instructor or division rule controls. Caltech also warns that detector errors can damage trust, privacy, and fairness.

Representative learning activity

Use the published Ethics and AI course case or a comparable first-year writing course. Preserve the initial human draft, any permitted AI exchange, disclosure, source verification, instructor conference, revision, and final reflection on voice and responsibility.

Protected human performance

Compare the final paper with an in-class analytical paragraph, oral conference, or proctored examination. The protected task should test argument structure, evidence, ethical reasoning, and the student's ability to recognize fabricated or misleading content.

Significance for the well formed mind

Caltech's strongest contribution is its protection of the relationship between trust and learning. Its rejection of detector-only enforcement avoids substituting an unreliable technical score for human judgment. The remaining question is whether the course designs improve durable performance.

COVID wellness and relationships

The public case record does not establish a comparable pre-COVID through AI-era outcome series. Review should include attention, trust, faculty contact, workload, and the effect of proctored or oral assessment on stress and belonging.

Evidence still needed

• Representative implementation outside Humanities and Social Sciences.
• Transfer from AI-supported writing to unaided scientific, mathematical, coding, and prose work.
• Student experience with disclosure, protected tools, proctoring, and detector restrictions.

Official sources Institute AI guidance | AI in teaching | HSS policy | Ethics and AI course

Case 6 Loyola Marymount University

Initial finding

LMU explicitly gives faculty practical options for AI-free and AI-integrated teaching. It also publishes a supported-tool list, requires attention to access and privacy before mandating a tool, and highlights a history-faculty example in which students help discuss and revise course expectations.

Published rule and governance

Faculty and departments determine academic use within honesty, security, and privacy requirements. Microsoft Copilot is available through LMU accounts with data protection. LMU tells instructors to consider account requirements, cost, accessibility, privacy, and a reasonable alternative before requiring student use.

Representative learning activity

Study a course that co-constructs its AI policy. Preserve the initial policy, student discussion, revisions, final rule, one AI-free assignment, one bounded-use assignment, disclosures, and the instructor's learning rationale.

Protected human performance

Use the AI-free assignment as the anchor and add a brief oral explanation or supervised revision. Students should identify their own reasoning, explain the policy tradeoffs, verify sources, and show that collaboration on the rule did not weaken individual accountability.

Significance for the well formed mind

LMU offers a useful model of agency because students may participate in setting expectations and because instructors are told to provide reasonable alternatives. The score depends on whether those choices protect foundational work and whether the university measures independent judgment.

COVID wellness and relationships

Whole-person education makes wellness relevant, but the case needs actual measures. Compare family and peer support, attention, workload, instructor contact, and help-seeking without treating current distress as proof of either COVID or AI effects.

Evidence still needed

• How widely co-constructed policies and AI-free alternatives are used.
• Outcome evidence for independent judgment, writing, and ethical reasoning.
• Local review of AI detection, accessibility, accommodation, and student data practices.

Official sources AI in teaching and learning | AI tools and services | Academic honesty and AI

Case 7 Thomas Aquinas College

Initial finding

The current California Code of Conduct resolves a major Phase Seven gap. Thomas Aquinas College does publish a student-facing AI rule. It states that AI can perform work students need to do for themselves and therefore prohibits AI or AI-enabled search engines in generating an academic paper. The prohibition extends to finding texts, outlining, obtaining arguments, composing sentences, editing, and AI-enabled tools such as Grammarly.

Published rule and governance

The rule is grounded in the College's educational purpose rather than detection convenience. Students are expected to learn from physical books, rational conversation, original texts, tutorials, laboratories, seminars, and composition. Special senior-thesis considerations are supplied separately to seniors.

Representative learning activity

Review one composition sequence and the senior thesis. Preserve reading notes, outlines, drafts, tutor comments, oral examinations or defenses, and the senior handout. Examine how sources are found and verified without AI-enabled search and how accessibility needs are handled.

Protected human performance

The ordinary program already contains protected performances: seminar discussion, mathematical demonstration, laboratory work, compositions, and defense of the senior thesis. Phase Eight should document scoring rubrics and the connection among these measures rather than invent a new test.

Significance for the well formed mind

This is the clearest institutional claim that some technologies conflict with the ends of education because they replace formative acts. It provides a strong benchmark for cognitive independence. It still requires evidence on implementation, accessibility, student learning, and whether graduates can later use AI responsibly when formation is secure.

COVID wellness and relationships

The College's emergency move online produced weaker engagement in its own account and reinforced the importance of in-person discussion. That experience supports institutional fit but is not a causal wellness study. Friendship, exercise, attention, and community participation should be examined as related conditions.

Evidence still needed

• The senior-thesis AI handout and any exceptions or accommodations.
• Implementation evidence and comparable outcomes before and after the explicit rule.
• How the College prepares students for ethical AI use outside academic-paper generation.

Official sources California Code of Conduct | College syllabus | AI and Human Flourishing | COVID online transition

Case 8 Mother of Divine Grace School

Initial finding

Mother of Divine Grace is evaluated only as a K through 12 program. Its strongest evidence concerns educational architecture: the curriculum follows primary, grammatical, dialectical, and rhetorical stages; parents hold an integral teaching role; consultants help individualize plans; and small online classes supplement the home program. No complete public generative-AI policy was located.

Published rule and governance

The absence of a public AI policy is recorded as an evidence gap, not permission or prohibition. Until a verified rule is available, the formation standard supports an H0 presumption for narration, memory, handwriting, arithmetic, initial composition, logical analysis, and other foundational tasks. Any AI use involving minors also requires age, consent, privacy, and parent-authority rules.

Representative learning activity

Begin with a grades 6 through 9 composition or logic sequence because it bridges learned foundations and independent argument. Preserve the lesson plan, parent role, student's notes and drafts, consultant or teacher feedback, oral discussion, grading rubric, and any digital tools used.

Protected human performance

Use the program's own human practices: oral narration, recitation, a handwritten or supervised composition, logical analysis, and discussion with a parent or teacher. The student should explain the argument and revise from human feedback before any optional AI comparison.

Significance for the well formed mind

MODG supplies the strongest family-mediated developmental model in the group. Its case can test whether a child reaches AI readiness through formed habits rather than early exposure. The unresolved policy question matters because age-specific boundaries and parent control are essential in K through 12.

COVID wellness and relationships

MODG entered 2020 with an established home-education structure and distinguished prepared homeschooling from crisis schooling. Its own account recognizes family stress, fear, isolation, and lost friendships. Comparisons must separate stable support from dysregulated family conditions and weak peer connection.

Evidence still needed

• Official AI rules for each developmental stage, approved tools, consent, data retention, and non-AI alternatives.
• Evidence showing how families currently handle AI-enabled search, grammar tools, tutors, and online classes.
• Comparable academic, friendship, family-burden, and wellness outcomes across delivery conditions.

Official sources School overview | Curriculum methodology | Curriculum stages | Crisis schooling account

Case 9 Harvard Westlake School

Initial finding

Harvard-Westlake's 2026 through 2027 curriculum guide includes a course on ethics, technology, and society in the age of artificial intelligence. The school also emphasizes independent thinking, integrity, interdisciplinary research, counseling, and wellness. Publicly accessible material does not yet establish one grades 7 through 12 AI-use framework.

Published rule and governance

Course-level expectations may exist behind student or parent portals, but they should not be inferred from the catalog. The case file must capture the actual syllabus, permitted tools, disclosure rule, age and consent requirements, privacy review, and the distinction between middle-school and upper-school use.

Representative learning activity

Study the AI ethics course through one problem-based inquiry. Preserve assigned readings, student research, any AI prompts, source verification, discussion, written argument, and final presentation. Add a comparison with a conventional middle-school writing task where AI is not used.

Protected human performance

Require an individually authored essay segment and an oral response to objections. A supervised source-evaluation exercise should test whether the student can detect unsupported claims, explain ethical principles, and distinguish personal judgment from generated language.

Significance for the well formed mind

The case can show whether studying AI ethically also forms the ability to resist convenient outsourcing. Because the school spans grades 7 through 12, a single rule would be inadequate unless it distinguishes developmental readiness and preserves parent notice and alternatives.

COVID wellness and relationships

The school's wellness and support system is visible, but a public longitudinal comparison was not located. Review should include friendship, belonging, sleep, workload, counseling, family expectations, and whether AI changes reliance on teachers or peers.

Evidence still needed

• The course syllabus, actual AI permissions, assignments, rubrics, and student work.
• Schoolwide or division-level rules for privacy, disclosure, age, consent, and academic integrity.
• Independent learning and wellness evidence by grade and prior support condition.

Official sources 2026 through 2027 curriculum guide | Independent research | Health and wellness

Case 10 High Tech High

Initial finding

High Tech High's board approved an Artificial Intelligence Use Policy on May 28, 2025 after a process involving feedback from students, staff, and the High Tech Education Collective. The discussion emphasized safety, optional teacher use, broad language that could be revised, handbook consequences, and resources for teachers. The public minutes establish adoption but do not reproduce the full policy.

Published rule and governance

The board record says the policy would be referenced in the student and staff handbooks and connected to consequences for violations. An earlier student presentation recommended reserving ChatGPT for higher grades, protecting fundamental grammar in elementary and middle school, and using AI as a guide rather than an answer. Those student recommendations are evidence of deliberation, not the final policy text.

Representative learning activity

Select one middle- or high-school project that culminates in a public exhibition. Preserve the project brief, individual and group roles, teacher's AI rule, prompts and outputs, drafts, critique, revision, community feedback, and the student's final presentation. Add an elementary foundational task as an age comparison.

Protected human performance

Inside collaborative work, require an individual process conference, oral defense, or supervised construction task. Each student should explain core knowledge, justify design choices, identify personal contributions, and respond to a new problem without AI.

Significance for the well formed mind

High Tech High is the strongest public K through 12 test of whether AI can support authentic projects without hiding weak individual mastery. The board process also creates a record of institutional authority and student and staff participation that can be compared with parent and public influence.

COVID wellness and relationships

Learning-recovery funding and post-COVID enrollment assumptions appear in board records. The case should connect recovery evidence to instruction only where data permit and should examine attendance, belonging, family burden, friendship, attention, and access to adult support.

Evidence still needed

• The complete adopted policy, revision history, approved tools, age rules, consent, privacy review, and annual review process.
• Evidence that each school and grade implements the policy consistently.
• Individual mastery, authorship, wellness, and equity outcomes inside AI-assisted group projects.

Official sources May 2025 board minutes | December 2023 board minutes | Design principles | Student and parent handbook

Cross case findings

Formation first The best case files begin with the mental act education intends to develop. A rule is defensible when it protects that act and when later assessment shows the learner can perform it independently.

Governance does not prove learning Licensed tools, committees, training, privacy terms, and course templates are necessary evidence of responsible administration. They do not establish memory, reasoning, judgment, authorship, or transfer after assistance ends.

Restriction can be educationally positive Thomas Aquinas College and Berkeley Law show that a prohibition may protect the work by which professional or liberal judgment develops. The justification, scope, accessibility, and independent outcome should still be examined.

Participation needs verification Stanford and High Tech High show the value of student voice, deliberation, collaboration, and authentic projects. Individual oral or supervised checks are still needed so strong group work does not conceal weak personal mastery.

K through 12 requires developmental rules Mother of Divine Grace, Harvard-Westlake, and High Tech High should be judged by age, parent authority, consent, privacy, foundational knowledge, and gradual transfer of responsibility. University criteria should not be imposed on children.

Wellness requires human evidence Attention, frustration tolerance, trust, belonging, friendship, family stability, workload, and help-seeking are relevant to formation. Policy pages cannot establish these outcomes. They require valid measures and careful comparison across pre-COVID, disruption, recovery, and AI-era conditions.

Phase Nine handoff

The next phase should convert these initial case files into an evidence ledger. For every institution, it should archive the exact policy and assignment record, mark each claim as verified, partial, missing, restricted, or superseded, and score the strength of the protected human-performance evidence. The first priority captures are the full High Tech High policy, Thomas Aquinas College's senior-thesis handout, Mother of Divine Grace's age-specific AI guidance if one exists, Harvard-Westlake's course syllabus and school rules, and implemented understanding checks at Stanford or comparable assessments elsewhere.