Issue #70 | Ungoverned at Work: The Jobs Shock Requires AI Minimum Viable Governance

Seventy issues in, AI governance is no longer only about tools, agents, or capability. It is about work, ownership, evidence, access, and who absorbs the consequences when AI scales.

AI Ethics & Governance for Leaders, Boards & Trustees

By Dr. Freddie Seba

© 2026 Freddie Seba. All rights reserved.

Editorial Note

This is Issue #70.

That number feels meaningful. Seventy issues ago, this newsletter began as a way to translate fast-moving AI developments into plain-language governance for leaders, boards, trustees, and institutional decision-makers.

Week by week, the same pattern kept appearing:

• AI was moving faster than policy.
• Adoption was moving faster than accountability.
• Vendors were moving faster than institutions.
• Experimentation was moving faster than evidence.
• Automation was moving faster than workforce redesign.
• Agents were moving faster than governance infrastructure.

Issue #69 framed this as the ungovernable moment: not because AI cannot be governed, but because many of the systems and institutions that govern technology were not designed for machine-speed action, multiplying workflows, agent identities, delegated authority, and decisions without audit trails.

This week, the urgency sharpened again.

The Economist argued that an AI jobs shock is not yet fully visible in labor-market data, but that governments should not wait for conclusive evidence before building safety nets. The same piece points to public anxiety, weak openings for college graduates, rising AI business spending, exploding agent use, and the risk that the quality and wages of future work may not be guaranteed. (mint)

Yale Insights framed the issue even more sharply: the real AI job destruction may be arriving before careers can start, with recent graduates finding it harder to get first jobs, early-career employment down in AI-exposed roles, and entry-level software roles under pressure. (Yale Insights)

That is the governance signal. AI is no longer only arriving as software.

It is arriving as:

• Capital allocation
• Workforce redesign
• Agentic infrastructure
• Financial services automation
• Clinical support
• Scientific acceleration
• Public-sector transparency challenge
• National security concern
• Mission-driven opportunity
• Institutional exposure

That is exactly the moment Ungoverned: A Practical Guide to AI Minimum Viable Governance was written for. Not because AI should be stopped. Because AI is already scaling. And comprehensive oversight often is not. https://www.amazon.com/clp/B0GY495GG1?lv=shuf

Market Signal

“Every engineer is a shipping agent. No one is governing them.”

The Guild.ai market language captures the moment almost perfectly:

Every engineer is a shipping agent. No one is governing them.

That sentence belongs in Issue #70. Not as an endorsement. As a signal.

Guild describes itself as a “control plane for AI agents,” focused on defining what agents can do, managing access to systems, and seeing what agents are doing. Its platform language emphasizes policies, permissions, observability, audit logs, and integrations across tools such as Slack, GitHub, Zendesk, Notion, Azure, Docker, GCP, and Google Docs. (Guild)

That is where the market is moving.

• From chatbots. To agents.
• From prompts. To workflows.
• From answers. To action.
• From the acceptable-use policy. To runtime governance.

A chatbot that gives a poor answer creates risk. An agent with permissions creates operational exposure. An autonomous workflow without auditability creates institutional fragility.

The question is no longer only: What model are we using?

The better questions are:

• What agents exist?
• What can they access?
• What actions can they take?
• What have they already done?
• What did they cost?
• Who approved their authority?
• Who owns the outcome?
• Who can stop them?

That is not future governance. That is now.

This Week’s Governance Lesson

The jobs shock requires AI Minimum Viable Governance.

The most urgent AI governance question this week is not simply whether AI will “replace jobs.” That question matters. But it is too narrow. The deeper governance question is: How is AI reorganizing work, authority, income, skill, dignity, accountability, and institutional responsibility before leaders can fully see the consequences?

• AI can change work before it eliminates work.
• It can change the review before it changes headcount.
• It can change performance expectations before job descriptions change.
• It can change staffing models before boards understand the risk.
• It can change entry-level pathways before universities, employers, and professional associations know how to respond.
• It can change the meaning of expertise before institutions define what must remain human-led.

That is why workforce AI belongs inside board governance. Not only as an HR side issue, but also:

• As a fiduciary issue.
• As a trust issue.
• As a people issue.
• As a long-term institutional legitimacy issue.

Executive Reflection

The Workforce Question Is Not Only Displacement. It Is Governance.

The workforce conversation is often framed as a prediction contest.

• Will AI eliminate jobs?
• How many?
• How fast?
• Which sectors?
• Which roles?

Those questions are important, but they are not enough.

Boards and executive teams need a different set of questions:

• Where is AI already changing work?
• Which roles are being redesigned informally?
• Which tasks are being automated without review?
• Which employees are bearing the new review burden?
• Which productivity gains are being captured, and by whom?
• Which workers have a voice in redesign?
• What happens to entry-level pathways when AI takes over first drafts, first reviews, first analyses, or first coding tasks?
• How will the organization distinguish augmentation from extraction?

Yale Insights describes a shift from task automation to workflow automation, in which people increasingly move from execution to exception handling, judgment, escalation, and oversight. It also reports a 16% decline in early-career employment across AI-exposed occupations since late 2022, with employment among developers aged 22 to 25 falling nearly 20% from its late-2022 peak. (Yale Insights)

That is not only a labor-market signal. It is a governance signal. It is because institutions are not simply buying tools. They are redesigning the future of work, and sometimes formally. Often quietly. And sometimes without saying so.

What We Are Seeing: Signals

1. The AI jobs shock is reaching the beginning of careers

The Economist’s warning is careful: the “AI jobs apocalypse” is not here yet, but waiting for conclusive evidence before creating a safety net may be too late. The article points to public anxiety, weak openings for graduates, rising business AI spending, data-center demand, and the risk that AI gains may flow disproportionately to owners of capital unless governments and institutions act early. (mint)

Yale Insights adds a higher-education and early-career lens: 41% of college presidents surveyed ahead of the Yale Higher Education Summit reported great concern about the vulnerability of entry-level white-collar roles, while only 10% said graduates were sufficiently or very well prepared for AI-enabled workplaces. (Yale Insights)

TechCrunch reported that Cisco is cutting nearly 4,000 jobs, or about 5% of its workforce, while also reporting better-than-expected profit and revenue and saying it is changing its cost structure to invest in AI and cybersecurity. (TechCrunch)

• AI governance translation: Workforce AI governance is not only about job loss. It is about job quality, entry-level pathways, worker voice, reskilling, accountability, and how productivity gains are distributed.
• Board/leader move: ask management for an AI workforce impact map before approving major AI-driven efficiency programs.
• The Ungoverned lesson: efficiency without accountability can become institutional fragility.

2. Agent governance is becoming platform governance

The agent signal is everywhere.

Guild.ai’s “control plane” language emphasizes permissions, observability, audit logs, and agent access management. (Guild)

Synthetic.ai describes an AI bookkeeping agent that connects to bank accounts, payroll systems, business tools, and email, then reads context, reconciles work, and asks clarifying questions. (Synthetic)

Anthropic’s Claude for Small Business integrates with tools such as QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365. It provides workflows for payroll planning, invoice chasing, campaign work, contracts, HR, customer service, and finance. Anthropic also states that users approve before something is sent, posted, or paid for, and that existing permissions carry over. (Anthropic)

This is useful. It is also governance-intensive.

Once AI connects to systems of record, governance must address identity, access, permissions, logging, spending authority, human approval, auditability, and shutdown.

• AI governance translation: once agents connect to institutional systems, governance must move from acceptable use to operational control.
• Board/leader move: require an enterprise agent inventory and access map before scaling agentic workflows.
• The Ungoverned lesson: if an AI agent can touch institutional systems, it is no longer just a productivity tool.

3. Delegated AI work is failing silently in long workflows

One of the most important research signals this week comes from the paper “LLMs Corrupt Your Documents When You Delegate.”

The researchers introduced DELEGATE-52, a benchmark that simulates long-delegated workflows across 52 professional domains. Their experiment with 19 models found that even frontier models corrupted an average of 25% of document content by the end of long workflows, with errors described as sparse, severe, silent, and compounding. (arXiv)

That matters far beyond document editing. Many institutional AI uses involve long workflows:

• Legal review
• Financial analysis
• Clinical documentation
• Policy drafting
• Grant writing
• Academic manuscript preparation
• Compliance documentation
• Code maintenance
• Procurement review
• Student support
• HR workflows

A system can perform well in a short task and still fail during extended delegation.

• AI governance translation: long-horizon delegation requires long-horizon evaluation.
• Board/leader move: do not approve consequential AI delegation based only on short demonstrations, vendor benchmarks, or isolated pilot tasks.
• The Ungoverned lesson: evidence matters more than fluency — especially when the workflow is long.

4. Clarification is a governance control

Another agentic research signal sharpens the same point. The paper “Ask Early, Ask Late, Ask Right” found that long-horizon agents can suffer cascading errors when they make a wrong assumption early in a task. Goal clarification loses nearly all value after the first 10% of execution. In contrast, input clarification retains value longer but declines sharply after the midpoint. The researchers also found that current frontier models do not reliably ask within the empirically optimal window. (arXiv)

That is a governance lesson hiding inside a technical paper. Institutions cannot assume agents will know when to ask. They need rules for asking.

• Before action.
• Before escalation.
• Before data exposure.
• Before irreversible workflow changes.
• Before spending.
• Before clinical, legal, financial, educational, or employment consequences.
• AI governance translation: clarification is not a courtesy. It is a control.
• Board/leader move: require “ask-before-action” rules for agents operating in consequential workflows.
• The Ungoverned lesson: human-in-the-loop is not enough if the human is consulted too late.

5. Agent architecture is governance architecture

The paper “Is Grep All You Need? How Agent Harnesses Reshape Agentic Search” shows that retrieval strategy, tool presentation, and agent-harness design can strongly shape agent performance, even when the underlying data remain the same. (arXiv)

The multi-agent paper “The Bystander Effect in Multi-Agent Reasoning” adds another warning: multi-agent collaboration does not automatically improve reasoning. The authors argue that unstructured multi-agent topologies can degrade independent reasoning and produce what they call “cognitive loafing” and “alignment hallucinations.” (arXiv)

The governance implication is straightforward: The system is not only the model. The system is the model plus: Harness + Tools + Retrieval design + Permissions + Data + Workflow + Logs + Users + Review + Escalation + Incentive aligment

• AI governance translation: tool architecture and multi-agent design are governance designs.
• Board/leader move: require AI reviews to include tool architecture, retrieval design, agent topology, permissions, logs, and workflow integration — not just model cards or vendor claims.
• The Ungoverned lesson: governance must cover the operating system around the model.

6. Financial services are scaling AI faster than supervisory capacity

The 2026 Global AI in Financial Services Report from Cambridge Judge Business School found that 81% of surveyed financial-services firms are adopting AI at some level, with 40% reporting advanced adoption. It also found that agentic AI is already in active adoption among 52% of industry respondents. (Cambridge Judge Business School)

The report identifies data privacy and protection, hallucinations and unreliable outputs, operational resilience, opacity, loss of human oversight, adversarial AI-related cyber threats, and bias/fairness as major risks. It also warns that agentic AI can scale cyber risk beyond the reach of manual oversight. (Cambridge Judge Business School)

The explainability gap is equally important: 78% of surveyed regulators rate explainability as critical or important, while only 50% of industry respondents have adopted explainable AI methods, and about two-thirds are not monitoring for bias or systemic discrimination. (Cambridge Judge Business School)

• AI governance translation: financial services leaders should treat AI as risk, conduct, trust, and fiduciary architecture — not only as productivity infrastructure.
• Board/leader move: require a financial-services AI control sheet that covers hallucination risk, recommendation boundaries, suitability assumptions, human override, escalation thresholds, audit trails, customer disclosure, explainability, and accountability for downstream harm.
• The Ungoverned lesson: confidence is not control, and fluency is not fiduciary judgment.

7. Health AI is moving quickly, but evidence and patient-data governance remain the test

Mayo Clinic announced that an AI model called REDMOD helped specialists detect pancreatic cancer on routine CT scans up to three years before clinical diagnosis; in a validation study, the model identified 73% of prediagnostic cancers at a median of about 16 months before diagnosis. (Mayo Clinic News Network)

That is promising, but it also reinforces the governance point. Health AI cannot be governed solely by performance claims. It requires evidence, validation context, clinical workflow readiness, patient data protection, false-positive and false-negative monitoring, escalation rules, and real-world implementation discipline.

The clinical regulation signal this week points in the same direction. The user-provided rheumatology article quote is worth keeping in the issue:

Because no single statute comprehensively governs clinical AI, institutions must proactively establish governance frameworks and invest in their implementation to ensure safe, transparent, and patient-centered use of AI in clinical care.

Reuters also reported that Britain’s NHS is granting staff from companies such as Palantir access to identifiable patient data via the National Data Integration Tenant platform. At the same time, NHS and Palantir statements emphasized strict policies, audits, approvals, and access controls. (Reuters)

• AI governance translation: Clinical AI governance cannot wait for statutory completeness or rely solely on vendor assurances.
• Board/leader move: require a clinical AI governance floor that covers evidence, data access, monitoring, human review, patient safety, procurement, vendor accountability, and post-deployment learning.
• The Ungoverned lesson: compliance is not the ceiling. In some areas, it is not even fully built yet.

8. Public-interest AI still needs public-interest governance

Anthropic and the Gates Foundation announced a four-year, $200 million partnership involving grant funding, Claude credits, and technical support for programs in global health, life sciences, education, and economic mobility. Anthropic says the work will include public health datasets, evaluation benchmarks, support for low- and middle-income countries, health-intelligence connectors, education tools, and agriculture-specific improvements. (Anthropic)

This is important. It shows how frontier AI may be directed toward public goods. But public-interest intention does not eliminate governance questions:

• Who defines the public good?
• Which communities are included?
• Which languages are supported?
• Who controls the tools?
• What data is collected?
• What benchmarks determine success?
• How are vulnerable populations protected?
• What happens when a public-interest workflow depends on proprietary infrastructure?

Such questions have real and immediate impacts on our organizations and society. It is up to all stakeholders to take agency as AI externalities have real consequences.

• AI governance translation: public-interest AI needs public-interest governance.
• Board/leader move: when AI partnerships serve vulnerable populations or public-good missions, they require governance for access, evaluation, language inclusion, community voice, sustainability, data rights, and risk of dependency.
• The Ungoverned lesson: good intentions do not replace governance.

9. Scientific and academic AI need disclosure infrastructure

A Springer Nature article calls for a unified checklist for reporting the use of LLMs in scientific manuscripts. The article argues that practical questions remain unresolved: what exactly should be disclosed, how disclosure should work, and how accountability can be promoted without stifling innovation. It proposes a checklist across conceptual contributions, linguistic assistance, and research assistance. It emphasizes that human oversight must involve subject-matter expertise and active verification, not merely a statement that a human was involved. The authors also suggest that future audit trails could include prompts, AI outputs, and version histories, especially for high-stakes publications. (Springer)

Australia’s Digital Transformation Agency offers a useful public-sector parallel: agencies must publish AI transparency statements that explain why they use AI, classify AI uses, identify cases where the public may interact with or be significantly impacted by AI without human review, describe monitoring and protections, and update statements at least annually or when material changes occur. (Digital Australia)

• AI governance translation: disclosure without specificity is weak governance.
• Board/leader move: require use-case-specific AI disclosure across research, policy, legal, clinical, student-facing, public-facing, and decision-support contexts.
• The Ungoverned lesson: transparency must be operational, not ceremonial.

10. AI claims are becoming liability claims

The Associated Press reported that Apple agreed to a $250 million settlement in a class-action lawsuit alleging false advertising of AI capabilities tied to Siri and Apple Intelligence. AP reported that the lawsuit alleged Apple promoted features that did not yet exist and misled buyers; Apple said it settled to resolve claims and stay focused on its products and services. (AP News)

This is important. It is a governance story. AI vendor claims need to be governed. Not just model behavior, marketing, or sales claims to use in promotion materials or procurement language, not performative bullet points for board decks, vendor demonstrations, or public statements.

• AI governance translation: AI assurance must be traceable to evidence.
• Board/leader move: require review of AI claims before they appear in marketing, procurement, board reporting, investor materials, or customer-facing communications.
• The Ungoverned lesson: assurance without proof creates reputational, legal, and trust risk.

11. Frontier AI risk management is dynamic, not one-time

The Oxford Martin AI Governance Initiative’s work on open problems in frontier AI risk management argues that most AI-specific risk management standards were developed for narrower AI systems and that frontier AI amplifies existing risks while introducing qualitatively new challenges. It identifies unresolved problems across risk planning, identification, analysis, evaluation, and mitigation. (Oxford Martin AIGI)

A related Oxford piece on determining the “state of the art” in general-purpose AI risk management argues that state-of-the-art risk management should be understood as a process-driven concept, established through scientific discourse and expert scrutiny, not provider practice alone. (Oxford Martin AIGI)

Reuters also reported that Microsoft, Google, and xAI agreed to give the U.S. government early access to AI models for national-security testing, allowing evaluation before deployment and research into capabilities and security risks. (Reuters)

• AI governance translation: Frontier AI risk management cannot be a one-time policy review.
• Board/leader move: require risk reviews to be updated whenever model capabilities, access, deployment context, regulation, or evidence change.
• The Ungoverned lesson: governance must learn at the speed of deployment.

12. Youth, mental health, and vulnerable users require upstream governance

The paper “AI and Suicide Prevention: A Cross-Sector Primer” states that AI chatbots already function as de facto mental-health support tools for millions of people, including people in crisis, while lacking the clinical validation, shared standards, and coordinated oversight their societal role demands. (arXiv)

That connects directly to a recurring theme in this newsletter: AI safety discussions often begin too late. Sometimes, they emerge after products scale, behaviors normalize, or after vulnerable users are already exposed. AI governance for vulnerable users cannot be added only after the harm is visible.

• AI governance translation: vulnerable-user governance must begin upstream, especially when AI systems simulate support, advice, companionship, therapy, coaching, or authority.
• Board/leader move: requires special review for AI tools that affect children, adolescents, patients, crisis users, students, workers, and other vulnerable populations.
• The Ungoverned lesson: protection is not a postscript. It is part of responsible design.

The Seba Framework

The 12 Ps of Responsible AI Oversight ©

This week’s signals fit clearly into the full Seba 12 Ps framework:

• Purpose — mission alignment versus convenience adoption
• Problems — what decision problem is actually being solved?
• Profits — who benefits versus who bears risk?
• People — patient, student, worker, customer, creator, and public impact
• Planet — compute, energy, infrastructure, and physical-world implications
• Process — monitoring, updates, escalation, rollback, and incident learning
• Policy — rules governing the use case and its limits
• Protections — red lines, vulnerable groups, and complaint pathways
• Privacy — data access, retention, exposure, masking, and secondary use
• Provenance — evidence, authorship, data lineage, model lineage, and traceability
• Preparedness — leadership competence, governance cadence, and operational readiness
• Product Ownership — who owns outcomes once AI shapes action?

Eight Ps feel especially active in Issue #70

People: Because AI is now changing work, entry-level opportunities, clinical care, education, research, mental health support, and public-interest systems.

• Profits: Because AI productivity gains, cost reductions, job cuts, capital allocation decisions, and market concentration raise questions about who benefits and who bears the risk.
• Process: Because agentic systems require access control, logging, monitoring, clarification rules, escalation, rollback, and runtime governance.
• Privacy: Because patient data, financial data, student data, business records, employee data, and public-sector information are increasingly connected to AI-enabled systems.
• Provenance: Because scientific writing, AI-assisted research, long delegated workflows, agent actions, and AI claims require traceability.
• Preparedness: Because institutions need governance before labor-market damage, clinical failures, agentic incidents, or public-trust failures become undeniable.
• Product Ownership: Because vendors may build the models and tools, but institutions own the context of use and the downstream consequences.
• Policy: Because workforce safety nets, clinical AI regulation, scientific disclosure norms, public-sector transparency, and frontier risk management are all policy questions, but institutional governance cannot wait for policy completeness.

Use Case

The AI Agent That Quietly Becomes a Worker

Imagine a company deploys an AI agent to help with operational work.

At first, it summarizes documents. Then it drafts responses. Then it connects to internal tools. Then it accesses email, tickets, code repositories, customer records, vendor invoices, and financial data. Then it starts routing work. Then it recommends decisions. Then the staff begins relying on it because it is fast, fluent, and always available.

However, no one calls it a worker. No one calls it a manager. No one calls it a fiduciary actor. No one calls it a system of record.

But functionally, it has started doing parts of all four. Then a problem emerges.

• A payment is mishandled.
• A customer receives the wrong response.
• A code change breaks production.
• A confidential record is summarized in the wrong environment.
• A job function is redesigned around the agent without review.

An employee is responsible for reviewing AI work, but without the time, training, or authority. A new graduate role disappears because the first-draft work has been automated. The institution looks for the governance record. There is a policy. There are vendor assurances. There may be meeting notes.

But no one can answer the operational questions:

• What was the agent doing?
• What was it allowed to do?
• What systems could it access?
• What did it actually do?
• What did it cost?
• What evidence supported delegation?
• What assumptions did it make?
• When did it ask for clarification?
• Who reviewed its work?
• Who owned the outcome?
• Who could stop it?

That is the gap. That is why agent governance and workforce governance now belong together.

Board-Ready Next Step

Require a 70-Day AI Work, Agent & Evidence Governance Sprint

For Issue #70, the board-ready action should match the milestone:

• Not a yearlong strategy project.
• Not an abstract AI committee.
• Not a policy refresh with no operating model.

A focused sprint designed to answer one question:

Where is AI already changing work, access, authority, evidence, and accountability — and what minimum governance floor must exist before it scales further? At a minimum, the sprint should produce seven outputs.

1. AI Work Impact Inventory

Identify where AI is already changing work.

Include:

• Task automation
• Drafting and review
• Coding
• Customer service
• Clinical documentation
• Financial analysis
• Legal review
• HR workflows
• Student support
• Research assistance
• Procurement
• Compliance documentation
• Shadow AI use

Do not ask only where AI is officially deployed. Ask where work has already changed.

2. Agent and Tool Access Map

Identify every AI-enabled system that can access institutional tools or data.

For each system, document:

• Systems accessed
• Data accessed
• Permissions granted
• Actions allowed
• Integrations
• Spending authority
• Human approval requirements
• Logs available
• Vendor dependencies
• Shutdown pathway

If the institution cannot see what AI can touch, it cannot govern what AI can do.

3. Workforce Redesign and Entry-Level Pathway Map

Document how AI is changing human work.

Ask:

• Which tasks are being removed?
• Which tasks are being added?
• Who reviews AI output?
• Who bears error risk?
• Who receives productivity gains?
• Which entry-level pathways are affected?
• Which roles are being quietly redefined?
• Which employees need training, time, authority, or protection?

AI governance is incomplete if it ignores people.

4. Delegation Readiness Test

Before delegating consequential work to AI, test beyond short demos.

Require evidence for:

• Long-horizon performance
• Error accumulation
• Silent degradation
• Clarification behavior
• Domain-specific reliability
• Auditability
• Human override
• Recovery from failure

A system that performs well on a short task may still fail when trusted with an extended workflow.

5. Clarification and Escalation Rules

Define when AI must ask before acting.

Especially before:

• Accessing sensitive data
• Making irreversible changes
• Spending money
• Modifying code
• Contacting customers or patients
• Producing formal institutional work product
• Making clinical, legal, financial, educational, or employment recommendations

Clarification is not a courtesy. It is a control.

6. Evidence, Disclosure, and Provenance Plan

For consequential AI use, document:

• What evidence supports deployment
• What disclosures are required
• What logs are retained
• What sources are traceable
• What AI contributed
• What human reviewed
• What assumptions remain uncertain
• What failures would trigger a review

This is how institutions move from assurance to proof.

7. Ownership and Stop Authority

For every consequential AI use case, name:

• Executive owner
• Operational owner
• Technical owner
• Legal or compliance owner
• Data owner
• Workforce owner
• Escalation owner
• Person or role authorized to pause or stop use

Distributed involvement is not accountability.

Ownership must be named.

Published Book Update

Ungoverned Is Now Available

My book, Ungoverned: A Practical Guide to AI Minimum Viable Governance, is now available on Amazon.

Book link: [Insert Amazon link]

This milestone issue is exactly why I wrote it.

The empirical foundation of the book comes from IRB-reviewed doctoral research at the University of San Francisco, grounded in interviews with faculty early adopters navigating AI in higher education, and extended through national convenings, including the AAC&U AI Institute and academic conferences hosted by AMIA.

It also draws from cross-sector practitioner conversations across higher education, healthcare, financial services, digital health, and technology — including the podcast and newsletter archive that helped validate the framework beyond a single institution.

AI governance is often discussed as if institutions must choose between two extremes: Move fast and accept the risk. Or: Wait until governance is perfect. I do not think either approach is sufficient.

The practical path is AI Minimum Viable Governance:

• Enough structure to prevent irresponsible normalization
• Enough clarity to assign ownership
• Enough evidence to support deployment
• Enough humility to admit uncertainty
• Enough authority to pause or stop use when conditions change
• Enough iteration to improve as reality changes

That is the governance floor.

And in this AI moment, the floor matters.

A Note to Readers

Why This Newsletter Is Getting Deeper

Some readers may have noticed that this newsletter has become more detailed. That is intentional. Like Ungoverned, this newsletter is designed for busy leaders who do not have time to read every AI article, track every technical paper, interpret every product announcement, or translate every market signal into governance action. That same “curate and translate” purpose was explicit in Issue #69 and continues here.

• The goal is not to overwhelm.
• The goal is to curate.

Each week, I try to identify what matters, explain why it matters, and translate it into questions leaders can use.

• A board member may focus on enterprise risk, ownership, and fiduciary oversight.
• A hospital leader may focus on safety, workflow, evidence, and trust.
• A university leader may focus on students, faculty burden, integrity, entry-level pathways, and mission.
• A startup founder may focus on product-market fit, pilots, trust, and responsible scaling.
• A financial-services leader may focus on explainability, conduct, fraud, cyber, and human oversight.
• A regulator may focus on public accountability and governance floors.
• A worker may focus on job quality, voice, fairness, and dignity.

That is the point. AI governance should not force leaders into simple binaries:

• Use or reject.
• Accelerate or slow down.
• Innovate or comply.
• Automate or protect people.

The better question is: What must be true for this technology to advance human flourishing in this specific context?

What I Am Watching This Week

• Whether boards begin asking for AI workforce impact maps, not just AI strategy updates.
• Whether companies treat AI-related job cuts as governance events rather than solely financial events.
• Whether agent platforms mature around identity, access, auditability, cost control, and shutdown authority.
• Whether long-horizon AI evaluation becomes standard before consequential delegation.
• Whether financial services close the gap between AI adoption and explainability, bias monitoring, and human oversight.
• Whether clinical AI governance moves faster than fragmented regulation.
• Whether patient-data access becomes a board-level governance issue rather than only an IT or privacy issue.
• Whether scientific AI adoption develops stronger provenance, disclosure, and audit trails.
• Whether public-interest AI partnerships strengthen access without creating new dependencies.
• Whether AI claims become more evidence-based before they reach customers, boards, or investors.
• Whether AI governance becomes a people capability, not just a compliance function.

The organizations that lead will not be the ones that adopt AI fastest. They will be the ones who can govern what AI changes.

Final Thought

Seventy Issues Later

Seventy issues later, the core argument is clearer to me than ever.

The question is no longer only: What can AI do?

The harder questions are:

• What work is AI changing?
• Who benefits?
• Who bears the risk?
• Who is displaced?
• Who reviews?
• Who decides?
• Who owns the outcome?
• Who can stop the system?
• What governance floor exists before scale?

The jobs apocalypse may not be here. But the governance test is. Waiting for perfect evidence before governing AI work is like waiting for the incident before building accountability.

The better path is harder — and more useful:

• Build
• Question
• Test
• Govern
• Listen
• Iterate
• Adapt
• Improve
• Start again

The goal is not to block AI’s benefits or diminish its potential. The goal is to make sure that potential serves people. That is the work of Ungoverned. That is the work of AI Minimum Viable Governance. And seventy issues later, 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 a global executive, Silicon Valley founder, and AI governance scholar-operator 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.

Global Executive | Silicon Valley Founder | AI Governance Scholar-Operator | MBA, Yale • MA, Stanford • EdD, USF

Gratitude + Mentions

Special appreciation to the readers, practitioners, board members, faculty, students, institutional leaders, podcast guests, and governance communities who have helped shape the first seventy issues of this newsletter. Special appreciation as well to the communities and institutions advancing responsible AI governance, health informatics, trustworthy implementation, workforce transition, public-interest AI, 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

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, or governance practices is intended for general understanding and should not 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 #AIJobs #AgenticAI #WorkforceAI #HealthAI #DigitalHealth #TrustworthyAI #AIandTrust #FinancialServicesAI #RiskManagement #HumanFlourishing #ProductOwnership #GovernanceAsCompetitiveAdvantage

Selected References Reviewed This Week

Links are provided for reader review and context. Inclusion does not imply endorsement.

Workforce, jobs, and transition

• The Economist / LiveMint — “Prepare for an AI jobs apocalypse.”
• Yale Insights — “The Real Job Destruction from AI Is Hitting Before Careers Can Start“
• TechCrunch — “Cisco cuts nearly 4,000 jobs to spend more on AI, reports ‘record quarterly revenue’“
• Partnership on AI — “Guidelines for AI and Shared Prosperity”

Enterprise agents, control planes, and delegated work

• Guild.ai — “The Control Plane for AI Agents”
• Synthetic.ai — “Autonomous accounting for startups.”
• Anthropic — “Introducing Claude for Small Business”
• arXiv — “LLMs Corrupt Your Documents When You Delegate.”
• arXiv — “Ask Early, Ask Late, Ask Right: When Does Clarification Timing Matter for Long-Horizon Agents?”
• arXiv — “Is Grep All You Need? How Agent Harnesses Reshape Agentic Search”
• arXiv — “The Bystander Effect in Multi-Agent Reasoning.”
• Better Path for AI — “Governance”

Financial services, risk, and supervision

• Cambridge Judge Business School — “2026 Global AI in Financial Services Report”
• AI Risk Navigator — “AI Risk Navigator”

Health AI, clinical governance, and patient data

• Mayo Clinic — “AI helps specialists detect pancreatic cancer up to 3 years before diagnosis.”
• Rheumatic Disease Clinics of North America — “Artificial Intelligence Regulation in the United States: Current Landscape and Implications for Rheumatology”
• medRxiv — “Shortkit-ML: A Unified Multi-Perspective Framework for Detecting Shortcut Learning in Medical Imaging Embeddings.”
• Reuters — “Britain’s NHS to grant Palantir contractors’ unlimited access’ to patient data, FT reports”
• DiMe Society — “Digital Health Measurement Product & Resources Navigator”

Public-interest AI, education, and vulnerable users

• Anthropic — “Anthropic forms $200 million partnership with the Gates Foundation”
• Student Guide to AI — “Human Wisdom for the Age of AI”
• arXiv — “AI and Suicide Prevention: A Cross-Sector Primer”

Scientific AI, academic integrity, and disclosure

• Springer Nature / Research Integrity and Peer Review — “A call for clarity: a unified checklist for reporting use of large language models in writing scientific manuscripts”
• arXiv — “AI co-mathematician: Accelerating mathematicians with agentic AI.”
• ScienceDirect — “The Cognitive Impact of ChatGPT in Higher Education”
• ADMS Centre — “GenAI Concepts”

Transparency, claims, and frontier risk management

• Australian Government Digital Transformation Agency — “Standard for AI transparency statements.”
• Associated Press — “Apple agrees to a $250 million payout after iPhone buyers allege Siri AI hype fell short.”
• Oxford Martin AI Governance Initiative — “Open Problems in Frontier AI Risk Management”
• Oxford Martin AI Governance Initiative — “Determining the State of the Art in General-Purpose AI Risk Management”
• Reuters — “Microsoft, Google and xAI to give US government early access to AI models for security checks.”