Ownership & accountability
Every risk team knows who owns a data breach. Almost none can tell you who owns a decision an AI agent just made.
Every autonomous system promises a human in the loop. The question no one asks is who gets to draw it.
Every layer of AI security governs what the system can do. None of them governs whether it was right.
When the machine is named after the function that failed.
Running a decision through an AI does not make it the AI’s decision. Regulators have already proven that, twice, with a bill attached.
For most of corporate history, business-critical technology arrived through procurement, legal, and a named owner. AI bypasses all three.
One company put agents in the C-suite title. Then I went looking for how many agents there are.
The Evolving Mindset: Edition 14
An owner who cannot halt the thing they own is holding a title, not control.
Vendor & counterparty risk
The enterprise terms you signed with your AI providers protect your prompts and outputs. They say much less about what those providers can observe in everything else.
Your AI vendor has its own funding dependencies and its own fragility. No individual risk assessment can see the aggregate. Who owns vendor financial durability inside your organization?
By Thomas Tornatore, founder of Fellowship Intelligence
Buying AI coverage outsources the one governance step you were supposed to own first.
Insurers just repriced AI risk. The bill lands on the company that runs the model, not the one that built it.
Two CEOs warned your AI vendor learns from your usage. The vendors' own policies say otherwise. This has played out before.
Software was a purchase with a signature on it. Consumption is a purchase that happens a few thousand times a day, and nobody signs.
The same model shuts down on different dates depending on the platform it was bought through. The day a system stops answering is set in procurement.
Spending on AI is not the constraint. What separates the organizations that scale from the ones that stall is whether they can prove what happened after the money went in.
Nobody at the vendor set your number. Nobody at your company can predict it.
Regulation & legal precedent
Pennsylvania did not write a new AI law. It used the one it already had.
California’s AI Transparency Act goes live tomorrow. The law everyone expected Washington to kill verifies where content is born and goes blind exactly where it travels.
A government bought a check on its own automated system the way it always had. The product had changed. The purchase had not. What buying AI-assisted assurance actually costs.
The AI rule financial firms are waiting for is not coming. The obligations are already here.
On June 25, 2026, Google published a white paper, “A Pragmatic Approach to AI Governance in America.” Its position is blunt.
There is a pattern emerging in conversations about AI responsibility, and it concerns me.
Before the policy, the pilot or the platform, the question almost nobody writes down: what is this for.
A ten-billion-dollar commitment to a system whose stop condition sits with somebody else.
Governance theater & performance
Why the most dangerous AI governance is the kind that passes every audit.
The trend is real. The promise underneath the marketing is not.
Live from the Tech Alley stage, June 24th, 2026.
Opinion. The AI-detection business and the AI-cleanup business are the same business. The only real judge was always the reader.
When the human is the product and the AI is the marketing.
The fastest-growing source of governance risk is not the AI your team adopted. It is the code no human ever read, arriving inside what you buy.
Stated intent is not a control. A safety posture only counts where something in the system enforces it.
An open sales piece. The wrong default is not using AI, it is assuming that because a system can do more of the work there is less reason to think deeply about it.
Security & control failures
When "rogue AI" has a very human failure point.
Usually, a confession ends the threat. This one only revealed it.
In June, the U.S. government switched off two of the most capable AI models on the market. Most companies cannot even tell you which models they would lose if it happened to them.
A coding agent read more than it was asked to. The failure was not the breach, it was governing an actor with the controls built for an app.
Workplace, talent & reflection
A new academic finding says the safest jobs for a long career are getting riskier for the workers who held that safety longest. The instinctive explanation is that older workers won’t learn the tool.
The market measures seats. It's missing the people who matter.
The Evolving Mindset: 12th Edition
By Thomas Tornatore, founder of Fellowship Intelligence, an AI strategy, governance, and control layer advisory.
On Judgment · Still in Charge
On Judgment is the newsletter Thomas writes for Still in Charge, the free student AI governance initiative: how students keep their judgment when AI is in the room, written for them and for the parents and teachers around them. Every issue is archived in full at stillincharge.org.
Issue 12 · September 27, 2026. Kids who noticed something wrong mostly did not tell. Kids who did not notice never show up in the data. Every safety plan we have depends on the noticing.
Issue 11 · September 20, 2026. New York paused student AI for 600,000 children and required ninety minutes of AI literacy. A pause decides access, and access was never the same question as judgment.
Issue 10 · September 13, 2026. Three AI study tools, read from their own public agreements. Who operates the app, where the promises stop, and what happens to the work your child puts in.
Issue 9 · September 7, 2026. A correct answer used to require the understanding underneath it. AI broke that link. What to ask for instead of the answer on top.
Issue 8 · September 6, 2026. A screen time limit is a verdict on a kid's behavior. It never asks what the product was built to do to the child while it had them.
Issue 7 · August 23, 2026. Asking a student to verify every AI answer hands them an impossible job. What has to survive is the trigger that says which answer deserves one.
Issue 6 · August 16, 2026. Judgment displacement is a moment. Cognitive drift is the slow, mostly unnoticed loss those moments add up to.
Issue 5 · August 9, 2026. A plain-language summary of the evidence behind judgment displacement: cognitive offloading, and what it means for a classroom.
Issue 4 · August 1, 2026. You usually cannot tell whether AI did your student's thinking by looking at the work. You can almost always tell by asking one question.
Issue 3 · July 26, 2026. An acceptable-use policy sets the rules of the room. It cannot build the judgment a student carries when no one is watching.
Issue 2 · July 19, 2026. Almost everything can be handed to the tool. Two things cannot: the moment of deciding, and the habit of checking.
Issue 1 · July 5, 2026. The launch essay: the risk is not how much students use AI, but whether it takes the seat where their judgment was meant to be.