Dr. Freddie Seba

AI Governance Keynote Speaker  ·  Author  ·  Scholar Operator

EdD · USF  ·  MBA · Yale  ·  MA · Stanford  · Teaching · UIC  ·  20+ years · Silicon Valley Founder & Global Executive · Digital Health · Fintech · Higher Ed.

Issue #60: Verification Is Governance Infrastructure Before Agency

March 9, 2026

Agentic AI, Clinical Oversight, Privacy Gaps, and the New Accountability Surface

AI Ethics & Governance for Leaders, Boards & Trustees

By Dr. Freddie Seba

© 2026 Freddie Seba. All rights reserved.

Issue #60 is a meaningful milestone, and I’m grateful to readers, guests, collaborators, and partners who have helped build this work with seriousness, trust, and shared purpose.

This Week’s Governance Lesson

AI capability is moving from generation to delegation.

But governance maturity only shows up when institutions build verification + oversight infrastructure before agency outruns accountability.

The institutional failure mode is now predictable:

If your institution expands an AI agency without a verification infrastructure, you do not get leverage.

You get speed → delegation → hidden error → escalation gaps → legitimacy loss.

So the board question becomes:

Do we have verification infrastructure and an accountability case… or are we scaling action faster than oversight?

Podcast Update

New episode drops midweek (Episode 9): Funding Meets AI Governance: Turning AI Hype Into Fundable, Defensible Companies.

Guest: Vijay Rajendran | Venture Builder, gAI Ventures | author of The Funding Framework | UC Berkeley Extension instructor

Why it belongs in this issue: verification is what turns AI enthusiasm into investable, governable execution. If leadership cannot clearly state the hypothesis, the success metrics, the test plan, and the stop/go criteria, then the company is not “early.” It is under-governed.

Available on @Spotify, @YouTube, @Apple Podcasts, and the @Substack app.

Executive Reflection: The Signal Is Not More Intelligent. It is More Delegated Judgment.

The signal this week is not that AI can answer more.

It is that AI is being positioned to recommend more, route more, document more, tutor more, and in some settings act more.

That creates a governance shift boards can’t ignore:

  • When judgment gets delegated, verification becomes the value layer.
  • And in high-stakes domains like healthcare, education, and financial services, verification becomes a safety layer rather than an administrative burden.

That is why this week’s signals matter. They point in the same direction:

agency is increasing, clinical deployment is getting closer to workflow reality, privacy risk is getting more ambient, and legitimacy is tightening around proof, provenance, and accountable human oversight.

That is not a call for panic. That is a call for preparedness.

What We are Seeing: Signals

1) Agentic AI shifts the control problem from content quality to delegated authority

A useful signal from OECD’s new framing of agentic AI: AI agents and agentic AI are not interchangeable. Agentic AI can coordinate multiple agents, decompose tasks, operate over longer periods, and function with limited human supervision in less predictable environments. Uptake is growing even as maturity in security, privacy, and robustness remains uneven. (OECD AI)

AI governance translation: Stop governing only outputs.

Start governing delegated authority:

  • What can the system access?
  • What can it recommend?
  • What can it trigger?
  • What can it change?
  • What can it do repeatedly without review?

Board move: Classify agency as a high-risk control tier requiring:

  • Explicit permissions
  • Human approval gates
  • Tool constraints
  • Memory/context rules
  • Rollback paths
  • Fast pause authority

2) Verified workflows are becoming a governance advantage—not just a productivity advantage

A strong technical signal from SkillsBench: the benchmark included 86 tasks across 11 domains, with 84 evaluated; curated skills raised average pass rates by 16.2 percentage points, gains varied widely by domain, and self-generated skills provided negligible or negative benefit on average. That is a governance lesson hiding inside a benchmark. (arXiv)

AI governance translation:

Institutions still need verified procedures, approved playbooks, and domain curation—not generic autonomy.

Board move: Require every high-impact AI use case to specify:

  • The approved workflow
  • The institutional playbook
  • The verifier
  • The success threshold
  • The no-go conditions
  • What must never be improvised

3) Clinical AI validation is local—not portable

A major clinical oversight signal from JAMA Network Open: a multicenter prospective validation of an updated proprietary sepsis prediction model across 227,091 inpatient encounters at 4 major US health systems found moderate to strong discrimination, but also high institutional variability, low positive predictive value, and high alert burden. The authors explicitly recommend local validation, workflow integration for false positives, and alert-silencing strategies. (JAMA Network)

AI governance translation:

Vendor performance is not deployment-grade assurance.

Board move: Require before deployment:

  • Local validation
  • False-positive management workflows
  • Alert-burden review
  • Threshold governance
  • Accountable operational owners

4) AI is becoming workforce-shaping infrastructure—not just task automation

A second clinical signal comes from recent experimental evidence in cancer diagnosis: in two preregistered field experiments with medical novices, AI input during training and during practice each improved diagnostic performance, and the combined condition produced the highest accuracy. The deeper governance lesson is not just that AI can assist. It is that AI can reshape the formation of capability itself.

AI governance translation:

If AI changes who can perform a task, it changes supervision design, credentialing assumptions, and escalation risk.

Board move: Require every expertise-adjacent AI deployment to declare:

  • Eligible user roles
  • Experience assumptions
  • Required supervision thresholds
  • Escalation rules
  • Override expectations
  • Where AI support ends, and licensed judgment begins

5) Patient-facing health AI remains a trust surface—not just a convenience surface

Signals from Yale’s clinician guidance point in the same direction: chatbots may improve access, translation, and health literacy, but anthropomorphization and conversational fluency can increase overtrust. That overtrust can lead to confusion, poor decisions, and mistrust in authentic care systems. Yale’s practical advice is clear: use chatbots more for information and efficiency than diagnosis. (Yale News)

AI governance translation:

Patient-facing AI should be governed as trust infrastructure.

Board move: For patient-facing or patient-influencing systems, require:

  • Intended-use boundaries
  • Do-not-diagnose / redirect protocols
  • Source expectations
  • Meaningful disclosure
  • Complaint-to-remediation pathways
  • Audit trails for recommendations, edits, and escalations

6) Evaluation is becoming the missing middle between demos and deployment

A recurring signal from healthcare governance work this month: there are still no settled benchmarks for many AI-enabled healthcare products, and developers and deployers are being warned not to rely on bold claims from unscientific evaluations. Post-deployment monitoring is not optional. It is part of the safety case. (Tech Policy Press)

AI governance translation:

A benchmark is a claim.

Deployment-grade evaluation is the verification system.

Board move: Require evaluation that matches the real setting:

  • In-workflow testing
  • Shadow mode or staged rollout
  • Task-specific error tolerance
  • Human factors review
  • Incident thresholds
  • Stop/go criteria tied to harm, not hype

7) Privacy gaps widen when AI becomes ambient

Two IAPP signals make the next governance trap very clear: consumer chatbot users may expect confidentiality that they are not actually receiving, and workplace transcription tools are creating consent gaps, retention problems, third-party processing exposure, and permanent records of sensitive conversations. Privacy risk is moving from discrete uploads to ambient capture. (IAPP.org)

AI governance translation:

Privacy cannot live in the footer.

It has to live in the workflow.

Board move: Require explicit governance for:

  • Recording and transcription defaults
  • Retention windows
  • Secondary-use restrictions
  • Sensitive-data segmentation
  • Storage and access boundaries
  • Notice and consent design

8) Oversight maturity is being built through exemplars, testbeds, and alignment capacity

A public-sector signal worth watching: the UK’s AI Exemplars programme says it is testing multiple AI solutions across real public-service use cases to learn quickly from both successes and failures while maintaining clinical oversight where needed. Separately, AISI says its Alignment Project announced 60 grant awardees and total funding of £27 million to advance research on systems that remain under human oversight and control. (GOV.UK)

AI governance translation:

Safety maturity is built through testbeds, evaluation capacity, and independent challenge—not just statements of principle.

Board move: Build some version of an internal safety inspector:

  • Pilot-to-policy pathways
  • Independent review capacity
  • Red-team or challenge functions
  • Post-deployment audits
  • Clear pause authority when systems outpace controls

9) Provenance and human authorship are hardening into legitimacy tests

Two governance signals on legitimacy now matter together. Reuters reports the U.S. Supreme Court declined to hear a case over copyright for AI-generated art, leaving intact the requirement of human authorship. Euractiv reports that researchers are struggling to find AI training data summaries that companies are supposed to provide under EU rules. Different contexts, same governance lesson: institutions will increasingly be judged on whether they can document what informed a system, how outputs should be used, and who remains accountable. (Reuters)

AI governance translation:

Provenance, disclosure, and accountability are moving from optional transparency to institutional defense.

Board move: Require provenance records for:

  • Vendor claims
  • Data and training disclosures
  • Output-labeling rules
  • Named accountable owners
  • Regulatory and litigation readiness

The Seba Framework: The 12 Ps of Responsible AI Oversight ©

How I translate signals into board-ready governance.

Purpose — mission alignment vs. cost extraction

Problems — decision-relevant framing (not metric chasing)

Profits — who benefits vs. who bears risk

People — patients/students/workforce; lived impacts

Planet — compute, infrastructure, scale costs

Process — lifecycle monitoring + incident learning

Policy — risk-specific rules (health, youth, education, employment)

Protections — vulnerable populations + escalation paths

Privacy — enforceable limits on data use/training

Provenance — traceability of data, models, vendors, outputs

Preparedness — board competence + governance cadence

Product Ownership — institutions own outcomes once AI acts

Board-Ready Next Step: A One-Page Agency & Verification Sheet

If you only do one thing this quarter:

Require an Agency & Verification Sheet for every AI use case. One page. Audit-ready. Updated post-deployment.

It should state:

Desicion Role (what does it do?)
  • Summaries
  • Recommendations
  • Triage/routing
  • Diagnostic support
  • Patient-facing explanations
  • Operational decisions
  • External-facing content

Action surface (what can it change?)

    • None (read-only)
    • Drafts only
    • Recommends
    • Routes / prioritizes
    • Sends with approval
    • Executes actions
    • Modifies records/triggers workflows

    Verification stack (3 layers)

      • Pre-action checks: approved sources, role access, tool boundaries, context limits
      • In-process structure: required rationale, citations, confidence markers, refusal rules, approval gates
      • Post-action review: sampling, peer review, audit logs, and incident learning

      Oversight case

        • Intended use + prohibited use
        • Eligible users + required supervision
        • Known failure modes (error, bias, overreliance, privacy leakage, context collapse)
        • Mitigations + accountable owners

        Quality + safety metrics (measured, not promised)

          • Task-specific accuracy/error estimates
          • Override rate
          • Escalation rate
          • Near-miss incidents
          • Time-to-correction
          • Equity/exclusion signals

          Monitoring + escalation

            • Who can pause
            • Who must be notified
            • What counts as an incident
            • Response timelines
            • Remediation + learning loop

            Value assessment

              • Time savings
              • Quality gains
              • Cost effects
              • Burden shift
              • Trust impact
              • Hidden exposure created or reduced

              This sheet turns “we’re piloting AI” into we are governing delegation.

              About the Author

              Freddie Seba is a researcher and practitioner—and an #AIKeynoteSpeaker and #AIEthicsSpeaker—focused on AI ethics and AI governance for leaders across higher education, healthcare, and financial services.

              He holds an MBA (@Yale University), an MA in International Policy Studies (@Stanford University), and an EdD in Organization and Leadership (@University of San Francisco), with a dissertation on AI ethics and governance defended in Fall 2025.

              He writes AI Ethics & Governance for Leaders, Boards & Trustees and hosts AI Governance with Dr. Freddie Seba, translating practitioner signals into board-ready oversight: decision rights, risk tiering, vendor accountability, monitoring, and incident preparedness.

              Speaking, Briefings, and Board Workshops

              Freddie Seba delivers a keynote on AI governance, risk, trust infrastructure, and institutional legitimacy. He also delivers AI executive workshops and board briefings with practical takeaways—accountability, transparency, safety, responsible adoption, escalation design, and governance maturity:

              Inventory → Tiering → Controls → Dashboards → Incident Drills

              To book: freddieseba.com and @LinkedIn.

              Gratitude

              Grateful for the communities that keep this work grounded:

              @University of San Francisco • @AMIA Informatics • @Stanford HAI • @Coalition for Health AI

              Transparency + Disclaimer

              Educational content only. This newsletter does not constitute legal, medical, clinical, insurance, financial, or professional advice.

              Drafted and refined with AI-assisted tools for synthesis and clarity. Final editorial control and responsibility remain with the author.

              © 2026 Freddie Seba. All rights reserved.

              #AIGovernance #ResponsibleAI #BoardOversight #AgenticAI #HealthcareAI #AILeadership #TrustInfrastructure #RiskManagement #AIandTrust #HigherEd

              SOURCES