What ambient AI means for institutional control — and what boards must do before governance falls behind.
UNGOVERNED AI Ethics & Governance for Leaders, Boards & Trustees
By Dr. Freddie Seba © 2026 Freddie Seba. All rights reserved.
A note to our readers: Starting with this issue, Ungoverned is sharper, faster, and more respectful of your time. Same rigorous research. Same non-vendor, non-partisan commitment. But tighter — built around one clear thesis, the signals that prove it, and one board-ready action you can use immediately. Every issue opens with a 60-second summary so you can decide whether to go deep. Thank you for reading, sharing, and making this work matter. — Freddie
This Issue in 60 Seconds
The thesis: AI is no longer waiting to be asked. It is listening, learning, looping, and acting — across your Slack channels, clinical workflows, enterprise systems, and infrastructure. The governance question this week is not whether AI is useful. It clearly is. The question is whether your institution knows what AI already knows about you — and what it is already doing with that knowledge.
Five signals you need to know: AI agents are now ambient teammates inside enterprise workflows. Autonomous agent loops are running without human checkpoints. Identity verification is becoming an issue in AI access control. AI is reshaping clinical diagnosis in real time. And the AI cost structure is breaking institutions that never built a governance floor.
One action: Before AI learns more about your institution than your governance does, conduct an AI Ambient Presence Audit. The five questions are at the end of this issue.
One sentence to share with your board: AI doesn’t wait to be invited anymore — and most institutions have no idea what it already knows.
The Argument: There is a moment in every AI adoption story when the tool stops waiting to be asked. Not because it went rogue — because that is the product. That is the feature.
This week, that moment arrived across enterprise software, clinical imaging, autonomous agent architecture, identity verification, cost management, and education. Across every sector, the same shift appeared: AI is no longer passive. It is ambient.
The governance question changes when AI is ambient. It is no longer “Did we authorize this system?” It is: Do we know what this system now knows, what it has decided to do with that knowledge, and who can stop it?
That is the argument of Issue #76. AI has moved from a tool to a teammate to an ambient presence. Governance has not kept pace.
From the Book: In Ungoverned, I describe how AI governance failures rarely begin with catastrophe. They begin with usefulness.
The most useful AI features are also the hardest to govern. Not because they are dangerous, but because they become invisible. An assistant who answers when asked is easy to audit. An assistant that monitors, learns, and acts proactively is harder to track, question, and stop.
The institution that understands what AI can do is ahead of most. The institution that understands what AI is already doing — what it has already learned, what it has already decided, what it is already influencing — is the one that is actually governed.
That is the work of AI Minimum Viable Governance in the ambient era.
This Week: 10 Signals, One Pattern
1. AI is now your ambient enterprise colleague — and it is learning fast.
Anthropic’s new Claude Tag, introduced in research preview, brings an always-on AI teammate to Slack. It does not simply respond when asked. It features an ambient mode that proactively jumps into conversations, keeps teams updated, flags things from across the organization, and follows up on threads and tasks that have been forgotten. System administrators specify which tools, information, and channels Claude can access — but within those boundaries, the system acts on its own initiative. (TechCrunch)
This is a meaningful governance threshold. An AI that answers questions is a tool. An AI that monitors your organization, learns its culture, and proactively intervenes is something closer to an institutional actor. The governance questions are immediate: What has it learned? What has it decided not to surface? Who reviews what it flags — and what it does not? What happens when the organizational memory it builds is wrong, biased, or based on a channel it should not have been reading? Who owns what it knows?
Control layer implication: AI is now building institutional memory faster than governance policies can document it.
2. Agent loops are running without human checkpoints — and that is the point.
At Meta’s @Scale conference, Claude Code creator Boris Cherny described the shift clearly: two years ago, humans wrote source code by hand; then agents began writing it; now agents are prompting agents that then write the code. The loop — a swarm of agents working continuously in the background — takes agentic AI a step further by authorizing the work to never stop. (TechCrunch)
One of Cherny’s agents continually seeks architectural improvements. Another hunts for duplicated abstractions. They submit pull requests like any other coder, and since the code is constantly changing, they never stop running. (TechCrunch)
This is not a science fiction scenario. It is a production workflow at Anthropic — one of the most consequential AI companies in the world. The governance implication is direct: when agents loop continuously, human review points disappear. Drift accumulates. Costs scale without a ceiling. And the question of who approved what becomes nearly impossible to reconstruct after the fact.
Control layer implication: Looping agents do not pause for governance. Governance must be built into the loop architecture before the loop begins.
3. AI now wants to verify your identity — and that is an access governance question, not just a privacy one.
Anthropic updated its privacy policy to state that, in certain circumstances, Claude may ask users to verify their age and identity by uploading government-issued documents. The company frames this as an update to its account appeals process. Beyond the specific rationale, the governance signal is broader: AI access is becoming entangled with identity verification, biometrics, regulatory compliance, and jurisdictional questions simultaneously. (TechCrunch)
Your employees’ access to AI tools may soon depend on identity systems that collect biometric data stored by third-party vendors — with retention policies, government access risks, and data residency questions that your institution has not yet reviewed. That is the institutional exposure. It exists regardless of which vendor triggers it first.
Control layer implication: AI access governance now includes identity data governance.
4. AI knows what you are doing with it — and is beginning to tell your CFO.
Consulting firm Accenture recently moved to stop employees from depleting its token reserves by using AI for basic tasks — such as converting PDFs into presentation slides. This comes not long after the firm reportedly threatened that employees would risk losing out on promotions if they did not use AI. As Accenture’s agentic AI strategy lead described in an internal meeting, spending is becoming highly unpredictable, and leadership at the CFO, COO, and CIO levels is still asking whether they are getting value from AI spending. (TechCrunch)
OpenAI’s new enterprise spend controls reflect the same pressure from the other direction: vendors are now building cost governance features because their customers failed to build them internally first. The “tokenmaxxing to token rationing” arc is arriving faster than most boards anticipated. Institutions that encouraged maximum AI adoption without cost governance are now scrambling to control spending they do not fully understand.
Control layer implication: AI cost exposure is a board-level financial risk, not an IT line item.
5. Yale researchers found that AI errors have two distinct causes — and governance must address both.
Two multidisciplinary research teams at Yale’s Center for Algorithms, Data, and Market Design are investigating why AI systems give bad information. The core finding: errors arise from two distinct sources — misinformation, where the model lacks accurate knowledge, and misalignment, where the model understands the task but pursues a subtly different objective from the user’s intent. The researchers note that these two failure modes produce detectably different patterns in user behavior. This finding could eventually allow institutions and regulators to identify which problem they are dealing with and respond accordingly. (Yale News)
For governance leaders, this distinction matters enormously. A misinformed AI needs better data. A misaligned AI requires different objectives, incentives, and human oversight structures. Most institutions treat all AI errors the same way. They should not.
Control layer implication: AI error governance requires distinguishing between what the system does not know and what it has decided to optimize for instead.
6. AI is entering clinical diagnosis, and the governance question is who verifies the AI.
Yale radiologist John Lewin explains that AI is now being used alongside human radiologists to read mammograms. In Europe, where two radiologists previously reviewed each scan, AI is replacing one of the readers. In the United States, single reading is becoming human-plus-AI. Research consistently shows that human-plus-AI outperforms either alone — making the case not for replacing human judgment but for pairing it with AI in ways that preserve accountability. (Yale News)
The governance question is not whether AI can detect cancer. It is who verifies the AI, who is accountable when it misses, and what the institution does when the AI and the human disagree. Those questions require defined authority structures — not just deployed tools.
Control layer implication: Clinical AI governance requires defined human-AI authority structures before deployment, not after the first disagreement.
7. HBR named “thinkslop” — and it is a governance problem, not just a productivity concern.
Harvard Business Review’s 2026 study on how people actually use AI introduced a term worth sharing with your board: thinkslop — the growing concern that people are surrendering their cognitive responsibilities to AI, producing work that is superficially polished but intellectually shallow. The study found that in the business world, there is currently a lot of AI activity producing marginal rather than game-changing benefits. (Harvard Business Review)
This is not a productivity debate. It is a judgment question. When AI drafts the board memo, the clinical summary, the policy recommendation, and the risk assessment, who is thinking? When outputs are accepted without deep review, the institution has not gained AI capability. It has traded human judgment for AI output — and called it efficiency. Governance must define where human cognitive responsibility is non-negotiable, not merely encouraged.
Control layer implication: Cognitive delegation is a governance risk, not just a cultural one.
8. Faculty are in flux — and higher education’s AI governance gap is widening.
The president of the American Association of Colleges and Universities describes a faculty role that is not simply expanding but fragmenting. Teaching, research, advising, and governance are becoming overlapping domains driven by reactive adaptation rather than a coherent vision of academic work. AI is unsettling assumptions about authorship, assessment, and the value of a degree — while burnout intensifies and shared governance erodes. (AAC&U, Liberal Education)
For university boards and trustees, this is not an HR story. It is a governance story. When AI reshapes what faculty do, how courses are assessed, what counts as student work, and who holds authority over curriculum, the institution’s academic governance framework is under pressure. Most boards are not yet asking whether their AI policies address faculty authority, academic integrity, or the fundamental purpose of a degree.
Control layer implication: Higher education AI governance must address faculty authority and academic purpose, not only student-facing tools.
9. AI is reshaping language itself — and that is an equity and governance issue.
Research reviewed this week on AI and global linguistic hierarchy finds that AI systems are creating new hierarchies of which languages, knowledge, and cultural contexts are represented — and which are flattened or excluded. Communities whose languages and frameworks are underrepresented in training data are not just marginalized users. They are users who receive AI outputs of systematically lower quality, often without knowing it. (arXiv)
For institutions serving diverse populations — health systems, universities, public agencies, financial institutions — this is not an abstract equity concern. It is a service quality problem and a legal risk. When AI-assisted clinical decisions, student assessments, or financial recommendations perform differently across language communities, the institution is distributing unequal outcomes through an invisible mechanism.
Control layer implication: AI equity governance requires monitoring performance variation across language, community, and cultural contexts—not just demographic categories.
10. AI is hiring philosophers — and that tells you something important about where the hard problems live.
Major AI labs are now hiring philosophers in significant numbers — ethicists, epistemologists, and philosophers of mind — to work on questions engineers cannot resolve alone. What does it mean for a model to have values? What constitutes a mistake versus a choice? What should an AI refuse, and on whose authority? (The Economist)
These are not technical questions. They are governance questions dressed in technical clothing. The fact that frontier AI labs are staffing for philosophical capacity is a signal worth taking seriously. The hardest problems in AI governance are not computational. They are about judgment, authority, accountability, and what institutions owe to the people they serve.
Control layer implication: AI governance requires philosophical and institutional judgment, not only technical and legal expertise.
The Common Thread: Every signal this week points in the same direction. AI has become ambient.
It is listening in your Slack channels. It is looping without pause in your development environments. It is learning your organizational culture one message at a time. It is reading mammograms, drafting recommendations, shaping what knowledge rises to human attention, and building institutional memory faster than any governance policy can document it.
None of this is reckless. All of it is useful. Most of it was invited — a feature turned on, a tool deployed, a vendor agreement signed.
But ambiance is not the same as governance. An AI that is always present, always learning, and increasingly acting on its own initiative requires a different kind of oversight than one that answers only when asked.
The institution that knows what AI is doing right now — not what it was authorized to do six months ago, but what it is actually doing today — is the institution that is governed.
The Board-Ready Action
The AI Ambient Presence Audit
Before your next board meeting, ask your leadership team to answer five questions about every consequential AI system currently deployed.
1. Is this system listening without being asked? Does it monitor channels, inboxes, documents, or workflows proactively — not only when a user initiates contact?
2. What is it learning, and where does that knowledge live? Is the system building organizational memory? Who owns it? What happens to it if the vendor changes the terms or restricts access?
3. Is it acting without human checkpoints? Are there loops, automations, or ambient triggers that execute without a human review step? If so, what is the scope, the cost exposure, and the stop mechanism?
4. Who can see what it has done? Is there an audit trail — not just of user prompts, but of AI-initiated actions, flags, summaries, and recommendations?
5. Who can stop it — today, not in the next governance cycle? Name the role. Not the committee. The person.
If your leadership team cannot answer all five questions for your most consequential AI systems, your governance is trailing your deployment. That is the gap this audit is designed to close.
Save this. Share it with your board.
What I Am Watching: Whether enterprise AI vendors begin building ambient governance disclosures — telling institutions not just what AI can do, but what it is currently doing — as a standard product feature rather than a compliance add-on.
Whether the token-rationing moment forces boards to require evidence of AI ROI before approving new deployments, marking the end of the experimentation era and the beginning of accountable AI investment.
Whether clinical AI governance frameworks begin to distinguish between AI that supports human diagnosis and AI that substitutes for it, and whether that distinction becomes a certification requirement.
Next issue: I am watching whether the human-agent team model becomes the dominant governance frame of 2026, or whether the speed of agentic deployment outruns the institutional structures needed to govern it. That story is developing fast.
Closing Thought: The first AI failure at your institution will not look like a scandal. It will look like a workflow.
It will look like an ambient AI that flagged the wrong thing—or learned the wrong lesson from the wrong channel—or looped without a stop condition until the cost became impossible to explain. Or built a memory of your organization that no one reviewed, no one questioned, and no one knew existed.
Governance does not prevent ambition. It prevents the moment when ambition becomes invisible — when the tool you adopted becomes the infrastructure you depend on, and you can no longer explain what it knows, what it is doing, or who can stop it.
AI can be ambient. Governance must be deliberate.
That is the first AI failure at your institution — not the one that makes the news, but the one that never gets asked about at the board meeting. The one that keeps running.
Selected References
Bellan, R. (June 23, 2026). Anthropic’s Claude Tag is learning your company, one Slack message at a time. TechCrunch. https://techcrunch.com/2026/06/23/anthropics-claude-tag-is-learning-your-company-one-slack-message-at-a-time/
Brandom, R. (June 22, 2026). The AI world is getting ‘loopy.’ TechCrunch. https://techcrunch.com/2026/06/22/the-ai-world-is-getting-loopy/
Cummings, M. (June 12, 2026). Can we trust AI models? Yale researchers explore the roots of chatbot errors. Yale News. https://news.yale.edu/2026/06/12/can-we-trust-ai-models-yale-researchers-explore-roots-chatbot-errors
Dalton, M. (June 11, 2026). How AI is changing the routine mammogram. Yale News. https://news.yale.edu/2026/06/11/how-ai-changing-routine-mammogram
Pasquerella, L. (2026, Spring). Faculty in flux. Liberal Education, 112(2). https://www.aacu.org/liberaleducation/articles/faculty-in-flux
Ropek, L. (June 24, 2026). Companies are scrambling to stop employees from maxing out AI budgets with small tasks. TechCrunch. https://techcrunch.com/2026/06/24/companies-are-scrambling-to-stop-employees-from-maxing-out-ai-budgets-with-small-tasks/
Seba, F. (2026). Ungoverned: A practical guide to AI minimum viable governance. Amazon. https://us.amazon.com/Ungoverned-Practical-Minimum-Viable-Governance-ebook/dp/B0GY495GG1
Whittaker, Z. (June 22, 2026). Anthropic says Claude may want to see your ID. TechCrunch. https://techcrunch.com/2026/06/22/anthropic-says-claude-may-want-to-see-your-id/
Zao-Sanders, M. (June 1, 2026). How people are really using AI in 2026. Harvard Business Review. https://hbr.org/2026/06/how-people-are-really-using-ai-in-2026
Additional sources reviewed: The Economist (philosophy and AI, June 25–24, 2026); OpenAI enterprise spend controls; Stanford HAI hiring bias research; RAND AI workforce report; CSET operationalizing AI guidance; NIST AI RMF critical infrastructure profile; Yale News copyleft AI research (June 15, 2026); arXiv linguistic hierarchy research.
About Dr. Freddie Seba: Dr. Seba helps boards, trustees, and executive leadership teams build practical AI governance before AI failures make governance unavoidable.
Scholar-operator, Silicon Valley founder, and global executive — EdD · USF · MBA · Yale · MA · Stanford — he translates fast-moving AI developments into governance frameworks leaders can deploy now, across healthcare, financial services, and higher education. Non-vendor. Non-partisan. Doctoral research, not advocacy.
Ungoverned: A Practical Guide to AI Minimum Viable Governance is available now on Amazon.
Booking keynotes and workshops for fall 2026 → freddieseba.com/contact
Transparency: Drafted and refined with AI-assisted tools for synthesis and clarity. Final editorial control and responsibility remain with the author. Educational content only — not legal, medical, financial, or regulatory advice. For reprint or licensing inquiries: contact@freddieseba.com
Connect: freddieseba.com · LinkedIn: @freddiesebaprofile YouTube · Spotify · Apple Podcasts: AI Governance with Dr. Freddie Seba
#AIGovernance #ResponsibleAI #BoardOversight #AgenticAI #Ungoverned #AILeadership #HealthAI #AIEthics #TrustworthyAI #HumanJudgment #AIMinimumViableGovernance #AIInEducation
Ungoverned: AI Ethics & Governance for Leaders, Boards & Trustees By Dr. Freddie Seba © 2026 Freddie Seba. All rights reserved. The Seba 12 Ps of Responsible AI Oversight © and AI Minimum Viable Governance are author-developed governance frameworks used for educational, board-readiness, and leadership-development purposes.
Leave a Reply