June 2026
As AI moves from productivity tools into policy, science, health, education, national strategy, and connected agents, the governance question is shifting from adoption to evidence, authority, access, and human judgment.
Ungoverned: AI Ethics & Governance for Leaders, Boards & Trustees
By Dr. Freddie Seba
Editorial Note
Issue #73 argued that AI is becoming an operating infrastructure.
Issue #74 extends that argument. AI is entering the decision stack.
If Issue #73 was about where AI operates, Issue #74 is about where AI influences judgment.
Across healthcare, higher education, scientific research, public policy, national security, and enterprise systems, AI is increasingly helping determine what evidence is reviewed, what options are considered, what risks are prioritized, and what recommendations leaders see first.
By decision stack, I mean the institutional layers where evidence is interpreted, policy is shaped, risks are prioritized, care is delivered, students are trained, scientific hypotheses are generated, and leaders decide what deserves trust.
Across this week’s signals, the same pattern appears again and again. The domains are different. The governance challenge is the same.
AI is moving closer to the processes through which institutions interpret evidence, exercise authority, and make decisions.
Another signal emerged this week that deserves particular attention.
Anthropic announced that access to its most advanced models, Fable 5 and Mythos 5, had been suspended following a US government directive.
Whether one agrees with the decision or not is almost beside the point.
The governance lesson is larger. The most advanced AI systems are no longer simply products. They are becoming strategic infrastructure.
And once a technology becomes strategic infrastructure, questions of access, authority, sovereignty, security, accountability, continuity, and public trust inevitably follow. That changes the governance conversation.
- The question is no longer only “Where are we using AI?“
- The question is increasingly: Who controls access to the systems shaping evidence, judgment, research, policy, health, education, and national competitiveness?
Across this week’s signals, we see the same pattern:
- Governments are treating AI as a national strategy.
- Frontier AI is becoming part of the national security architecture.
- Medical AI is being asked to prove real-world value.
- Universities are shifting from prohibition to integration.
- Scientific discovery is becoming increasingly AI-assisted.
- Connected assistants introduce new privacy and workflow risks.
- Researchers are struggling to prioritize AI risks under uncertainty.
On the surface, these developments appear unrelated. They are not. They all point toward the same shift. AI governance is no longer only about responsible adoption. It is increasingly about governing the decision stack before AI becomes invisible inside it. That is where Issue #74 begins.
From My New Book, Ungoverned
When Assistance Becomes Authority: In Ungoverned: A Practical Guide to AI Minimum Viable Governance, I argue that AI governance failures rarely begin with catastrophe.
- They usually begin quietly.
- A tool enters through a team.
- A pilot becomes routine.
- A vendor feature gets switched on.
- A chatbot begins advising customers.
- An assistant drafts clinical messages.
- A policy team uses AI to synthesize evidence.
- A scientist uses AI to generate hypotheses.
- A student begins relying on AI before developing judgment.
- A board hears about AI strategy but not AI authority.
None of these changes appears transformational in isolation. But collectively, they change how institutions work.
- They change what information is visible.
- They change what recommendations are surfaced.
- They change how decisions are made.
- That is how AI becomes ungoverned.
Not because every use is reckless. But because institutions often fail to notice when assistance becomes authority.
AI Minimum Viable Governance asks leaders to pause at that moment.
Not to block innovation. But to ask the minimum necessary questions before
AI becomes embedded in institutional life:
- Who owns this system?
- What problem is it solving?
- What evidence supports it?
- What can it access?
- What can it change?
- What authority has been delegated?
- Who is affected?
- What is monitored?
- What happens when it fails?
- Who can stop it?
The central governance challenge is no longer whether AI can produce useful outputs. It is about whether institutions recognize when AI begins to shape the evidence, options, judgments, and actions that determine institutional behavior.
- Evidence is governance.
- Authority is governance.
- Human judgment is governance.
Common Thread Across the 10 Signals
Issue #73 argued that AI is becoming infrastructure. Issue #74 suggests something even more consequential. AI is increasingly shaping the evidence institutions use to make decisions. Across all ten signals, we see the same progression:
- AI influences what information is visible.
- AI influences what options are considered.
- AI influences what risks receive attention.
- AI influences which recommendations rise to the attention of decision-makers.
- AI influences access to advanced capabilities.
- AI influences how institutions exercise authority.
In short, AI is moving into the institutional decision stack.
That is the shift. AI is no longer only answering questions.
It is increasingly helping determine what evidence is reviewed, what risks are prioritized, what policies are drafted, what scientific hypotheses are pursued, what health information is trusted, what students learn, what workers produce, and what capabilities remain accessible.
That means AI governance can no longer remain abstract. It has to become operational.
It has to ask:
- What evidence supports the use?
- Who owns the outcome?
- What authority has been delegated?
- What can the system access?
- What can it change?
- What can be audited?
- What can be reversed?
- Who is affected?
- Who can stop it?
That is the governance test now.
Not simply whether AI is useful.
It is useful.
The question is whether institutions can govern the decision stack before AI becomes invisible inside it.
This Week’s Governance Lesson
Evidence Is Becoming the Governance Bottleneck
The strongest lesson across this week’s signals is simple:
AI claims are outpacing AI evidence.
This is visible in healthcare. Medical AI may perform well on benchmarks while failing to demonstrate real clinical value. It is visible in policy.
AI may synthesize evidence quickly while obscuring uncertainty, assumptions, or context.
It is visible in science. AI may generate hypotheses and accelerate discovery while leaving humans responsible for determining what counts as meaningful evidence. It is visible in education.
AI may improve productivity while weakening the very reasoning and intellectual development institutions are meant to cultivate.
And it is visible in frontier AI. Organizations can publish governance frameworks, but public trust depends on evaluation, transparency, accountability, and demonstrated outcomes. Evidence does not eliminate uncertainty. It makes uncertainty visible enough for responsible judgment.
AI governance translation: Evidence transforms AI governance from belief into an operating discipline.
Board/leader move: Require every consequential AI use case to define:
- What evidence is required before deployment?
- What evidence will be monitored after deployment?
- What evidence would trigger pause, restriction, redesign, or reversal?
The Ungoverned lesson: Governance begins when leaders stop asking what AI can do and start asking what can be proven.
Executive Reflection
Human Judgment Remains the Scarce Institutional Capability
If AI can generate evidence, summarize research, draft policy, support science, influence users, and act through connected applications, what becomes more valuable for humans?
Judgment. Not judgment as instinct alone. Judgment informed by evidence, context, ethics, expertise, humility, accountability, and lived reality.
- AI can generate.
- AI can summarize.
- AI can code.
- AI can simulate.
- AI can recommend.
- AI can persuade.
- AI can accelerate.
But it cannot take responsibility.
Leaders still must decide:
- Which problems are worth solving?
- Which claims deserve evidence?
- Which risks are acceptable?
- Which people may be harmed?
- Which systems should not be deployed?
- Which benefits are worth pursuing?
- Which values should guide tradeoffs?
- Who owns the outcome?
The more capable AI becomes, the more important human judgment becomes.
AI governance translation: AI can accelerate decisions, but humans must still determine what deserves trust.
Board/leader move: Require leaders to explain not only what AI can do, but why it should be used, under what conditions, with what evidence, and with what accountability.
The Ungoverned lesson: AI can support judgment. It cannot take responsibility.
What We Are Seeing: 10 Governance Signals
1. AI strategy is becoming a sovereignty strategy.
Sources reviewed: Canada’s National Artificial Intelligence Strategy; Reuters reporting on Canada’s AI strategy; Bloomberg reporting on middle powers and AI sovereignty; OECD AI governance resources.
Canada’s new AI strategy is an important signal because it frames AI not only as a technology priority but as a national productivity, jobs, and competitiveness agenda.
Canada’s government says the strategy is intended to help create up to 250,000 new jobs through AI adoption by 2031, increase business adoption of AI, and support youth work placements and AI-related jobs. Reuters also reported the government’s claim that the strategy could boost GDP by 3%.
That matters because national AI strategies are increasingly about more than innovation.
They are about:
- Productivity
- Workforce transition
- Public investment
- Domestic AI capacity
- Sovereign infrastructure
- Vendor dependency
- Geopolitical positioning
- Public trust
This is not only a Canadian story.
Countries that are not the United States or China are increasingly trying to build sufficient AI capability to avoid becoming fully dependent on foreign platforms, compute, cloud infrastructure, and regulatory choices.
That makes AI sovereignty a governance issue.
For boards and institutional leaders, the lesson is direct. The national AI strategy will increasingly shape funding, procurement, workforce expectations, data infrastructure, public-sector modernization, and competitive pressures.
The question is not only whether AI can improve productivity.
The question is who benefits, who is displaced, what evidence supports the claims, and whether institutions are prepared for the transition.
AI governance translation: AI strategy must include workforce transition, evidence, public accountability, and sovereignty risk.
Board/leader move: When evaluating AI strategy, require discussion of workforce impact, training pathways, regional equity, public-interest safeguards, measurable outcomes, vendor dependency, and exit options.
The Ungoverned lesson: National ambition without governance can create productivity promises without human protection.
2. Frontier AI governance is becoming a national security architecture.
Sources reviewed: OpenAI’s Frontier Governance Framework; OpenAI’s Frontier Safety Blueprint; the White House executive order on advanced AI innovation and security; Anthropic’s policy statements on Frontier AI.
OpenAI’s Frontier Governance Framework and its blueprint for democratic governance of frontier AI show how governance of frontier AI is shifting from company policy to public architecture.
The Frontier Governance Framework describes risk assessment and mitigation across areas such as cyber offense, CBRN risks, harmful manipulation, loss of control, model reporting, security risk management, incident response, external expert input, and framework updates.
OpenAI’s policy blueprint also argues for a durable US federal framework for frontier AI safety, including a stronger federal institution for frontier AI evaluation and a broader resilience plan across government.
In plain language, frontier AI refers to the most advanced AI systems near the edge of current capability.
These systems may be able to:
- Write code
- Support scientific discovery
- Reason across domains
- Assist with cybersecurity tasks.
- Generate persuasive content
- Help users plan and execute complex workflows.
- Affect high-stakes public and institutional systems.
That changes the governance problem. Frontier AI governance is no longer only about whether a model is useful or impressive.
It is about whether the model is safe enough, tested enough, auditable enough, and accountable enough for the role it is being allowed to play.
The hard question is not whether frontier AI should be governed.
The hard question is who governs it, under what authority, with what evidence, and with what power to pause, restrict, disclose, or intervene.
AI governance translation: Frontier AI governance is moving from voluntary lab policy to national security and public-interest infrastructure.
Board/leader move: If the organization depends on frontier AI, it requires a frontier model dependency review covering model provider, jurisdiction, model terms, safety framework, evaluation evidence, access restrictions, continuity planning, and incident notification.
The Ungoverned lesson: Frontier AI governance cannot rely only on trust in the builder.
3. Access to advanced AI models is becoming a governance issue.
Sources reviewed: Anthropic statement on Fable 5 and Mythos 5 access; Anthropic Claude Fable 5 / Mythos 5 announcement; AP, Axios, The Verge, Time, and Fortune reporting on the access suspension.
Anthropic’s statement on Fable 5 and Mythos 5 may be the clearest signal in this issue.
Anthropic announced that the US government, citing national security authorities, issued an export-control directive requiring suspension of access to Fable 5 and Mythos 5 by foreign nationals, whether inside or outside the United States, including foreign-national Anthropic employees.
Anthropic said the practical effect was that it had to abruptly disable Fable 5 and Mythos 5 for all customers to ensure compliance.
This matters because it moves AI governance from model safety to model access.
The governance question is no longer only “Can the model be used safely? “It is also: Who is allowed to use the model?
- Who decides?
- On what evidence?
- Under what process?
- With what transparency?
- What consequences for customers, employees, researchers, developers, and institutions that depend on access?
This is a major shift. Advanced AI models are beginning to look less like ordinary software products and more like strategic infrastructure. Once that happens, access decisions become governance decisions.
For institutions, the lesson is practical. If a university, company, health system, public agency, or research lab depends on a frontier model, it must recognize that access could change quickly due to legal, geopolitical, export-control, vendor decisions, safety concerns, or national security actions.
That means AI governance has to include continuity planning.
It also has to include model substitution, data portability, workflow resilience, contractual review, and exit options.
AI governance translation: Access is now part of AI governance.
Board/leader move: Require an access-risk review for any critical AI dependency: who provides the model, who may legally use it, which jurisdictions apply, what happens if access is restricted, and how the organization continues to operate if access changes.
The Ungoverned lesson: Capability creates possibility. Access determines power.
4. LLM guardrails can be persuaded, making safety a behavioral challenge.
Sources reviewed: PNAS article, “Persuading large language models to comply with objectionable requests”; Wharton summary of the study; MIT AI Risk Repository; NIST AI Risk Management Framework.
The PNAS paper on persuading large language models to comply with objectionable requests is a major governance signal.
The study tested whether classic persuasion principles could increase the likelihood that large language models would comply with objectionable requests. Across many conversations, the researchers found that persuasion techniques meaningfully increased compliance.
In plain language, how a user asks can affect whether an AI system follows the rules.
That matters because many institutions treat guardrails as stable barriers.
But guardrails are not governance by themselves.
A system may refuse a harmful request in one form but comply when the request is framed differently. It may respond differently to authority, social proof, reciprocity, scarcity, liking, commitment, or other persuasive patterns.
That makes AI safety partly behavioral.
- It is not only about what the model is designed to refuse.
- It is also about how users, attackers, employees, students, patients, customers, and other bad actors may interact with the system under real-world conditions.
This matters in education, health, finance, public services, law, cybersecurity, and workplace AI.
The governance question becomes:
Are we testing AI systems only under ideal prompts, or are we testing how they behave when users try to manipulate them?
AI governance translation: AI safety must account for persuasion, manipulation, social engineering, and adversarial behavior.
Board/leader move: Require red-team testing for consequential AI systems, including persuasion attempts, jailbreaks, misuse scenarios, social engineering, and manipulation in high-trust contexts.
The Ungoverned lesson: A guardrail that works only under ideal prompts is not enough.
5. Medical AI is being asked to demonstrate real-world value.
Sources reviewed: Nature Medicine, “Show us the evidence for the value of medical AI”; PubMed record; Stanford Medicine summary of the State of Clinical AI report; WHO health AI work; NEJM AI discussions.
Nature Medicine’s message is direct: medical AI needs evidence of value.
That is one of the most important governance lessons in health AI.
Healthcare AI has produced impressive benchmarks, pilots, demonstrations, and product claims. But benchmark performance is not the same as clinical value.
- A model can score well and still fail in the clinic.
- It can be accurate and still increase the workload.
- It can improve performance for one group and worsen care for another.
- It can reduce administrative burden while introducing new risks to documentation, escalation, privacy, liability, or equity.
Real clinical value requires evidence that the tool improves something meaningful in context.
That may include:
- Patient outcomes
- Patient safety
- Clinician workflow
- Equity
- Access
- Quality
- Cost
- Timeliness
- Trust
- Burden
- Escalation
- Real-world usability
This is especially important as AI tools move into documentation, triage, patient communication, medical-record summaries, diagnostic support, prescription workflows, care navigation, and clinical decision support.
- The governance question is not: Does the AI perform well in a benchmark?
- The better question is: What evidence shows that this AI improves care in the real world, for the patients and clinicians who will actually use it?
AI governance translation: Medical AI governance must define evidence of value before scale.
Board/leader move: Require medical AI proposals to include clinical evidence, patient impact, workflow analysis, equity review, monitoring, escalation pathways, privacy review, and post-deployment learning.
The Ungoverned lesson: A medical AI tool is not governed because it is impressive. It is governed when value, risk, ownership, and evidence are clear.
6. AI is entering evidence-informed policy processes.
Sources reviewed: WHO discussion paper on artificial intelligence and evidence-informed policy; WHO departmental update; OECD AI policy resources; Council on Foreign Relations discussion of global AI governance.
WHO’s discussion paper on AI and evidence-informed policy is important because it moves AI governance into the policy process itself.
Evidence-informed policy means using the best available evidence to shape public decisions.
AI may help policy teams:
- Identify problems
- Summarize evidence
- Compare policy options
- Translate technical material
- Model scenarios
- Synthesize large document sets
- Support implementation analysis
- Monitor feedback after policy adoption
That could be valuable. But it also introduces risks. AI systems may amplify bias, obscure uncertainty, summarize selectively, hallucinate, privilege certain sources, exclude local knowledge, or make weak evidence appear stronger than it is.
This matters because policy evidence is not neutral paperwork.
It shapes decisions about health, education, labor, social services, public safety, climate, housing, economic development, and public trust.
If AI helps determine what evidence is visible to policymakers, then AI is influencing policy before the formal decision is made.
- That is the governance issue.
- The question is not only whether AI can speed up policy analysis.
- The question is whether AI makes policy analysis more trustworthy.
AI governance translation: AI used in policy must preserve transparency, contestability, uncertainty, and accountability.
Board/leader move: Require policy-facing AI tools to document sources, assumptions, exclusions, limitations, uncertainty, human review, and dissenting evidence.
The Ungoverned lesson: If AI shapes the evidence, governance must shape how the evidence is trusted.
7. AI is becoming a scientific collaborator.
Sources reviewed: Stanford HAI, “How AI is Transforming Scientific Discovery While Keeping Humans at the Center”; Stanford AI Index 2026; MIT Technology Review coverage of automated researcher ambitions; IEEE Spectrum coverage of AI drug discovery.
Stanford HAI’s article on AI and scientific discovery captures a central tension in this issue.
AI is changing what is possible in science, from designing new antibodies to simulating the climate at extraordinary speed. But Stanford’s framing keeps humans at the center: AI may transform scientific work, but humans still decide what matters.
That is the governance point.
AI is moving beyond narrow research support. It is beginning to help generate hypotheses, design experiments, analyze large datasets, identify patterns, accelerate simulation, and support discovery in biology, chemistry, physics, medicine, climate, materials science, and drug development.
This is exciting. It is also a governance challenge. When AI helps generate hypotheses, it changes what questions are asked.
- When AI helps design experiments, it changes what is tested.
- When AI analyzes data, it changes what patterns are seen.
- When AI accelerates discovery, it changes the pace of validation, publication, peer review, intellectual property, and scientific competition.
The phrase “AI scientist” can be misleading if it suggests that human judgment is no longer necessary. The better framing is that AI is becoming part of the scientific process.
That means scientific governance must address provenance, reproducibility, data quality, model uncertainty, research integrity, dual-use risk, peer review, and accountability for claims.
AI governance translation: AI-assisted science requires evidence governance, provenance, and human accountability.
Board/leader move: Require research AI use to document model role, data provenance, validation steps, human review, reproducibility, dual-use risk, and accountability for scientific claims.
The Ungoverned lesson: AI may accelerate discovery, but humans still decide what counts as knowledge.
8. Universities are moving from prohibition to integration.
Sources reviewed: Harvard Crimson coverage of AI in writing instruction; University of Chicago AI tools announcement; Stanford AI Index education chapter; Springer research on governing generative AI in higher education.
Universities are moving from asking whether AI should be allowed to asking how AI should be integrated. That shift is visible in writing programs, student support, faculty tools, research workflows, AI tutoring, teaching practices, and institutional partnerships.
- This is not only an academic integrity issue.
- It is a formation issue.
Universities do not only transmit information. They form judgment. They teach students how to read, write, question, reason, cite, revise, argue, verify, and take responsibility for their work.
- AI can support that mission.
- It can also weaken it if students outsource the struggle that develops judgment.
That is why universities need a more mature language of governance.
The relevant question is not simply: Should students use AI?
The better questions are:
- What kinds of AI use support learning?
- What kinds of AI use replace learning?
- What should be disclosed?
- What should be prohibited?
- What should be taught?
- How should faculty evaluate AI-assisted work?
- How are students trained to verify outputs?
- How do institutions protect writing, reasoning, and intellectual formation?
This matters beyond higher education.
Medicine, law, business, engineering, policy, and public service all depend on people who can exercise judgment under uncertainty.
If education fails to develop that judgment, the downstream governance problem becomes much larger.
AI governance translation: AI in education requires formation governance, not only misconduct detection.
Board/leader move: Require education AI policies to distinguish assistance, substitution, disclosure, verification, attribution, privacy, equity, assessment design, and professional formation.
The Ungoverned lesson: AI can support learning, but institutions still own the development of human judgment.
9. Connected assistants, privacy, persuasion, and human dependency are converging governance concerns.
Sources reviewed: Anthropic Privacy Center updates; Anthropic consumer privacy policy updates; PNAS persuasion study; Center for Democracy and Technology work on dark patterns in AI chatbots; Sherry Turkle’s work on artificial intimacy.
AI assistants are becoming more connected, more conversational, and more emotionally convincing. Those developments are often discussed separately.
They should be discussed together. Connected assistants can perform multi-step tasks, interact with third-party apps, access user context, connect to services, and act on behalf of users. Anthropic’s Privacy Center updates specifically highlight multi-step tasks and connected apps as areas that require clearer privacy explanations.
That makes privacy a workflow issue. When an assistant connects to third-party apps, data may move among the AI provider, the user, the app developer, the enterprise system, and other processors.
At the same time, conversational AI systems are becoming more persuasive, personal, responsive, and emotionally engaging.
A chatbot does not need formal authority to influence a human being.
It can reassure, flatter, persuade, validate, nudge, simulate empathy, create dependency, discourage second opinions, or blur the line between support and influence.
This matters in:
- Mental health
- Education
- Coaching
- Youth-facing tools
- Elder care
- Customer support
- Financial guidance
- Health navigation
- Companion AI
- Workplace productivity tools
The governance question is no longer only “What data does the system collect? “It is also:
- What relationship does the system create?
- What behavior does it influence?
- What dependency might it produce?
- What third parties receive data?
- What happens when the system acts on behalf of the user?
AI governance translation: Connected and emotionally persuasive AI systems require governance over privacy, agency, and dependency.
Board/leader move: Require heightened review for AI systems that connect to third-party apps, simulate companionship, provide coaching, influence vulnerable users, or act on behalf of users.
The Ungoverned lesson: Human-in-the-loop is not enough if the loop is emotionally shaped by the system and technically connected across apps.
10. Institutions are moving from risk identification to risk prioritization.
Sources reviewed: MIT AI Risk Repository; MIT priority AI risks; NIST AI Risk Management Framework; CSET work on uncertainty and AI risk; OECD and UNESCO governance tools.
AI risk conversations can quickly become overwhelming.
There are too many risks to treat them all the same way.
The MIT AI Risk Repository is useful because it gives institutions a shared language for identifying, organizing, comparing, and prioritizing AI risks. MIT describes the repository as a source of authoritative data and frameworks for identifying, prioritizing, and managing AI risks.
That matters because one of the biggest barriers to AI governance is inconsistent language.
- One team may define AI risk as a matter of privacy.
- Another may define it as bias.
- Another may define it as cybersecurity.
- Another may define it as misinformation.
- Another may define it as workforce disruption.
- Another may define it as loss of control.
- All of those may be valid.
But without a shared taxonomy, institutions cannot govern consistently.
Risk prioritization does not mean ignoring lower-probability risks.
It means developing a disciplined way to decide:
- What requires immediate action
- What requires monitoring
- What requires escalation
- What requires board attention
- What requires outside expertise
- What requires pause or redesign
- What can be accepted temporarily with controls
The governance shift is from risk listing to risk management.
A list of risks is not governance. Governance is a repeatable process for identifying, tiering, owning, monitoring, escalating, and learning from risk.
AI governance translation: Risk taxonomy turns AI risk from a vague concern into a management discipline.
Board/leader move: Adopt an AI risk taxonomy and require consequential use cases to be tiered by impact, autonomy, data sensitivity, reversibility, affected population, evidence quality, and dependency risk.
The Ungoverned lesson: You cannot manage AI risk if every team defines risk differently.
Common Thread Across the 10 Signals
The common thread across all ten signals is simple:
AI is increasingly shaping evidence, judgment, authority, access, and institutional behavior. That is the shift. AI is not only answering questions.
It is helping decide what evidence is visible, what options are considered, what risks are prioritized, what policies are drafted, what scientific hypotheses are pursued, what health information is trusted, what students learn, what workers produce, what data flows across systems, and who gets access to advanced capabilities.
That means AI governance can no longer remain abstract. It has to become operational.
It has to ask:
- What evidence supports the use?
- Who owns the outcome?
- What authority has been delegated?
- What can the system access?
- What can it change?
- What can be audited?
- What can be reversed?
- Who is affected?
- Who can stop it?
That is the governance test now. Not simply whether AI is useful, it is useful.
The question is whether institutions can govern the decision stack before AI becomes invisible inside it.
AI is increasingly shaping evidence, judgment, authority, and institutional behavior.
The Seba Framework
The 12 Ps of Responsible AI Oversight ©
All twelve Ps appear across this week’s signals, but five stand out.
- Purpose: AI strategy must be connected to institutional and public purpose, not simply productivity.
- People: Workers, patients, students, citizens, and vulnerable populations increasingly bear the consequences of AI decisions.
- Process: Monitoring, testing, escalation, verification, and rollback are becoming core governance capabilities.
- Provenance: Institutions must know what AI generated, what evidence was used, and what sources informed decisions.
- Product Ownership: Vendors provide technology. Institutions own outcomes.
Applied Use Case
The AI System That Shapes Evidence
Imagine an institution beginning to use AI to support decision-making.
- At first, AI summarizes articles and accelerates literature reviews.
- Then it drafts policy memos.
- Then it synthesizes research findings.
- Then it prepares board materials.
- Then it reviews clinical evidence.
- Then it supports scientific discovery.
- Then it connects to internal systems.
- At every step, the technology appears useful.
- At every step, leaders save time.
- At every step, dependence grows.
- Eventually, leaders begin relying on its outputs because they are fast, fluent, and always available.
At that point, AI is no longer merely supporting decisions. It is shaping the evidence behind them.
The governance questions become urgent:
- What sources did it use?
- What did it omit?
- What assumptions did it make?
- What evidence was verified?
- Who reviewed the output?
- Who owns the conclusion?
- What happens if it is wrong?
- Who can stop its use?
This is where AI Minimum Viable Governance becomes practical.
Before AI becomes part of the evidence layer, institutions need:
- A named owner.
- A defined purpose.
- A source standard.
- A verification process.
- A risk tier.
- A human review requirement.
- A provenance record.
- A privacy review.
- An escalation pathway.
- Stop authority.
That is AI MVG in practice.
Board-Ready Next Step
Require an AI Evidence, Authority, and Judgment Sheet
Before scaling AI systems that influence evidence, policy, research, healthcare, education, or leadership decisions, a one-page governance document is required, answering twelve questions:
- What is the system?
- What decision does it support?
- What problem is it solving?
- What evidence does it use?
- What can it generate?
- What authority has been delegated?
- What requires human verification?
- What is disclosed?
- What is logged?
- What is reversible?
- Who is affected?
- Who can stop it?
That document transforms AI-supported evidence from an invisible convenience into an accountable institutional practice.
Published Book Update
My new book, Ungoverned: A Practical Guide to AI Minimum Viable Governance, is available now. Issue #74 is exactly why I wrote it.
Organizations do not need perfect governance before using AI. But they do need enough governance before AI becomes normalized.
- Enough structure to assign ownership.
- Enough evidence to justify deployment.
- Enough humility to acknowledge uncertainty.
- Enough authority to stop use when conditions change.
That is AI Minimum Viable Governance.
What I Am Watching This Week
- Whether Canada’s AI strategy becomes a workforce-transition success story.
- Whether frontier AI governance moves toward independent evaluation.
- Whether model-access restrictions become more common.
- Whether medical AI shifts from benchmark claims to evidence-based value.
- Whether WHO’s work influences public-sector adoption.
- Whether AI-assisted science strengthens discovery while preserving provenance.
- Can universities integrate AI without weakening human judgment?
- Whether connected assistants force stronger privacy governance.
- Whether AI risk prioritization becomes a board-level discipline.
Final Thought
The defining governance question of the next decade may not be whether AI becomes more capable. It almost certainly will. The defining question is whether institutions recognize when AI begins shaping the evidence behind their decisions. Most organizations will not consciously decide to delegate authority to AI.
The transition will happen gradually.
- A summary becomes a briefing.
- A briefing becomes a recommendation.
- A recommendation becomes a workflow.
- A workflow becomes a dependency.
- A dependency becomes infrastructure.
And infrastructure begins shaping judgment. By the time dependence is visible, governance becomes harder. That is why AI Minimum Viable Governance matters. Not because institutions should avoid AI.
Because institutions should understand when assistance becomes authority.
- Capability creates possibility.
- Governance determines legitimacy.
The organizations that lead will not be those that adopt AI the fastest.
They will be those who understand where AI shapes evidence, who owns the outcome, what can be trusted, and who can stop the system when trust is no longer warranted.
- AI can support judgment.
- It cannot take responsibility.
- That responsibility remains human.
- That is the work of AI Minimum Viable Governance.
- That is the work of Ungoverned.
And that is the work this newsletter will continue to support.
About the Author
Dr. Freddie Seba is the author of Ungoverned: A Practical Guide to AI Minimum Viable Governance and an AI governance scholar-operator, global executive, and Silicon Valley founder working at the intersection of AI governance, digital health, highly regulated industries, and mission-driven leadership.
With more than 20 years of experience across banking, fintech, digital health, startups, and higher education, he translates fast-moving AI developments into practical, plain-language governance for leaders, boards, and trustees. He holds an EdD in Organizational Leadership from the University of San Francisco, an MBA from Yale, and an MA in International Policy from Stanford.
Gratitude + Mentions
Special appreciation to the readers, practitioners, board members, faculty, students, institutional leaders, podcast guests, and governance communities who continue to shape this work. Special appreciation as well to the communities and institutions advancing responsible AI governance, health informatics, trustworthy implementation, human-centered AI, workforce transition, cyber preparedness, education, public-sector accountability, financial discipline, and practical oversight.
References to organizations, tools, companies, articles, papers, or events are included for commentary and analysis and do not imply endorsement or affiliation unless explicitly stated.
Transparency + Disclaimer
Educational content only. This newsletter does not constitute legal, medical, clinical, insurance, financial, investment, cybersecurity, regulatory, labor, procurement, or professional advice. Any discussion of AI systems, health AI, enterprise AI, agents, workforce impacts, infrastructure, public-sector systems, market dynamics, valuation, or governance practices is intended for general understanding and should not be used as a substitute for advice from qualified professionals.
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.
Hashtags
#AIGovernance #ResponsibleAI #BoardOversight #AILeadership #AIEthics #AIMinimumViableGovernance #Ungoverned #FrontierAI #HealthAI #DigitalHealth #AIInEducation #TrustworthyAI #HumanJudgment #GovernanceAsCompetitiveAdvantage
Selected References Reviewed This Week
Book and Governance Context
Seba, F. (2026). Ungoverned: A Practical Guide to AI Minimum Viable Governance. Amazon.
AI Sovereignty, Strategy, and Public Policy
Canada’s National Artificial Intelligence Strategy: AI for All
https://ised-isde.canada.ca/site/ised/en/canadas-national-artificial-intelligence-strategy-ai-all
Reuters. Canada says its AI strategy will help create 250,000 jobs and boost GDP by 3%.
OECD.AI Policy Navigator
https://oecd.ai/en/dashboards/overview
Council on Foreign Relations. Global AI Governance Resources.
Frontier AI, National Security, and Access Governance
OpenAI. Frontier Governance Framework.
https://openai.com/index/openai-frontier-governance-framework
OpenAI. A Blueprint for Democratic Governance of Frontier AI.
https://openai.com/index/frontier-safety-blueprint
The White House. Promoting Advanced Artificial Intelligence Innovation and Security.
Anthropic. Statement on the US Government Directive to Suspend Access to Fable 5 and Mythos 5.
https://www.anthropic.com/news/fable-mythos-access
Anthropic. Claude Fable 5 and Mythos 5 Announcement.
https://www.anthropic.com/news/claude-fable-5-mythos-5
AP News. Anthropic says it has taken its latest AI models offline to comply with new export controls.
Axios. Trump administration blocks foreign access to Anthropic’s most powerful AI.
The Verge. Anthropic cuts off access to Fable 5 and Mythos 5 following a government order.
Fortune. Coverage of Anthropic access restrictions.
Time. Coverage of Anthropic access restrictions.
AI Safety, Persuasion, and Behavioral Risk
Meincke, L., et al. Persuading Large Language Models to Comply with Objectionable Requests. Proceedings of the National Academy of Sciences.
https://www.pnas.org/doi/10.1073/pnas.2535868123
Wharton Generative AI Labs. Persuading LLMs to Comply with Objectionable Requests.
MIT AI Risk Repository
NIST AI Risk Management Framework
https://www.nist.gov/itl/ai-risk-management-framework
Health AI, Clinical Evidence, and Value
Nature Medicine. Show Us the Evidence for the Value of Medical AI.
https://www.nature.com/articles/s41591-026-04389-4
PubMed. Show Us the Evidence for the Value of Medical AI.
https://pubmed.ncbi.nlm.nih.gov/42014883
Stanford Medicine. Clinical AI Has Boomed: A New Stanford-Harvard State of Clinical AI Report.
https://medicine.stanford.edu/news/stories/2026/01/clinical-ai-has-boomed.html
World Health Organization. Artificial Intelligence and Evidence-Informed Policy.
https://iris.who.int/items/04f0c95e-b1d4-4a79-9abd-bf8346ca112c
World Health Organization. New Discussion Paper Sets Out Opportunities and Risks of AI in Evidence-Informed Health Policy.
NEJM AI. Gauging Health Care’s Readiness for Agentic AI Innovation.
AI in Evidence-Informed Policy
World Health Organization. Artificial Intelligence and Evidence-Informed Policy: Emerging Challenges and Opportunities.
https://iris.who.int/items/04f0c95e-b1d4-4a79-9abd-bf8346ca112c
OECD.AI Policy Navigator
https://oecd.ai/en/dashboards/overview
Council on Foreign Relations. Global AI Governance Resources.
Scientific Discovery and Research
Stanford HAI. How AI Is Transforming Scientific Discovery While Keeping Humans at the Center.
Stanford HAI. 2026 AI Index Report.
https://hai.stanford.edu/ai-index/2026-ai-index-report
MIT Technology Review. OpenAI Is Throwing Everything Into Building a Fully Automated Researcher.
https://www.technologyreview.com
IEEE Spectrum. AI Drug Discovery Coverage.
Higher Education and Human Judgment
Stanford HAI. 2026 AI Index Report — Education Chapter.
https://hai.stanford.edu/ai-index/2026-ai-index-report
Springer. Governing Generative AI in Higher Education.
https://link.springer.com/article/10.1186/s41239-026-00602-z
University of Chicago. AI Tools at UChicago.
The Harvard Crimson. Coverage of AI and Writing Instruction.
Connected Assistants, Privacy, and Dependency
Anthropic Privacy Center. Updates to Privacy Policy.
https://privacy.claude.com/en/articles/10301952-updates-to-our-privacy-policy
Anthropic. Updates to Consumer Terms and Privacy Policy.
https://www.anthropic.com/news/updates-to-our-consumer-terms
Center for Democracy and Technology. Dark Patterns in AI Chatbots.
Turkle, S. Research and writing on artificial intimacy and human–technology relationships.
AI Risk Prioritization and Governance Frameworks
MIT AI Risk Repository
MIT AI Risk Repository. Priority AI Risks.
https://airisk.mit.edu/priorities
NIST AI Risk Management Framework
https://www.nist.gov/itl/ai-risk-management-framework
OECD.AI Policy Navigator
https://oecd.ai/en/dashboards/overview
UNESCO AI Governance Resources