01Student
11:40pm Sunday. Jayden's essay is due first period. The approved AI tool only signs in on his school Chromebook, which is in his locker. His phone is on the desk, already logged into ChatGPT.
Gateway AI in EDU Summit / 23 July 2026
AI Governance Is Not
a Policy Document.
The distance between the rules a district writes and what actually happens the moment a student or staff member hits Enter.
Caleb Vail Founder, TenetInteractive working sessionApprox. 50 minutes
A show of hands
You wish the rule
was enough.
Who here has kids? You want them to follow every rule, and to never cross the street alone. And none of us leaves it at the rule. We hold the hand, we watch the corner, we build the habit until it holds without us. Your AI policy is the rule you wrote down. Your students and staff are already at the street.
What a written rule cannot touch
Here is what is actually at stake
Advance to reveal each83%
could not quote their own essay
At stake · The ability to think
They stop building the muscle
In an MIT brain-scan study, students who wrote with ChatGPT showed the lowest brain engagement of any group, and 83% could not quote a single line of the essay they had just finished. Students feel it themselves: two-thirds now say the more students use AI, the more it harms their critical thinking.
MIT Media Lab, Your Brain on ChatGPT, 2025 · RAND American Youth Panel, 20261 in 50
children hit every year
At stake · Their data privacy
A child's record is the prize
A student's data is a clean slate: an untouched Social Security number and spotless credit. That is why children are targeted, about one in fifty every year, and why the theft stays hidden for years, often until they apply for a first loan, job, or apartment. Whatever gets pasted into an ungoverned tool does not come back.
Javelin Strategy, Child Identity Fraud, 2021 to 20221 in 3
teens: AI over a person
At stake · Learning to be with people
The friend that only ever agrees
AI companions are always available and built to keep kids talking, which means built to agree. Among the 72% of teens who have tried one, a third have already turned to it instead of a real person for a serious conversation. For a generation still learning how to be with each other, a friend that never pushes back is its own kind of risk.
Common Sense Media, Talk, Trust, and Trade-Offs, 2025None of this is settled by the document in the binder. It is settled the moment someone hits Enter. So let us look at how fast that moment is arriving.
The problem
AI is already everywhere. The policy is the only thing standing still.
Students and staff use AI every day, on district devices and their own. One in ten teens now do all or most of their schoolwork with a chatbot. Most districts answered with a written policy, and almost none can say whether it changes anything the moment someone opens a chat.
Teens using AI chatbots for schoolwork
13%
26%
54%
202320242025
Doubled, then doubled again. Pew Research, 2023 to 2026. 2023–24: ChatGPT specifically; 2025: any AI chatbot.
6 in 10
teachers used AI for their work this school year.
Gallup / Walton Family Foundation, 202582%
of teachers have received no formal guidance on using AI.
Gallup, 2026The plan
How this session works
Part 1
The tradeoffs
The real tensions in writing an AI policy, and the nuance most rules ignore.
Part 2
The test
In groups of three, we run a reasonable policy through a normal school day, live.
Part 3
The break, and the cost
Where these policies fail, how districts try to enforce them, and what that enforcement costs in reliability and in people.
This is not only about students
Governance has two arenas
Exhibit 1Arena 1
The classroom
Students using AI for the work itself: academic integrity, safety, and the personal data they paste into a prompt.
Arena 2
The back office
Staff and the program you are building: teachers, counselors, and administrators putting rosters, IEPs, and records into AI, plus the internal tools a district stands up on its own data.
A program that covers one arena and forgets the other is not a program. Most policies forget the second.
Self-assessment
Three levels of AI control
Exhibit 2 / click your levelMost districts live at Level 1, believe they are at Level 2, and need Level 3.
Before we write a single rule
Good policy has to hold real tensions
For discussionTeacher autonomyvsDistrict consistency
Teachers advocate for their own judgment and academic freedom, and the best classroom use is often the one a teacher invented. A blanket rule can override the professional closest to the student.
UsablevsSafe
Too restrictive and students leave for ungoverned tools. Too open and the guardrails are gone. The rule has to keep AI worth using on the governed path.
OversightvsPrivacy
To govern use you have to see it, and seeing it creates its own surveillance risk. Governance has to act without hoarding what students and staff type.
The open question
Should the rules change by grade? A second grader and a graduating senior are not the same user, yet most policies treat them as one.
K-56-89-12
Working session
Test a policy against
a normal day.
Find two people near you. Each takes one role and plays it honestly, pressures and all. The goal is not to design a perfect policy. It is to watch a reasonable one meet reality.
Groups of three / assign now
The three roles
Protects the district / builds the program
The Administrator
"I answer for everyone, and I cannot be everywhere."
- Owns FERPA, COPPA, and reputation
- Is standing up the internal AI program
- Has authority, but no view into the chat
Protects the learning
The Teacher
"I want them to think, not outsource the thinking."
- Cares about learning and integrity
- Also a daily AI user with student data
- Little visibility into what is typed
Protects their time
The Student
"I am not trying to cheat. I am trying to be done."
- Path of least resistance wins
- Has a phone, home wifi, personal login
- Treats friction as a thing to route around
How it works
Three short rounds
1
Write good rules · 5 min
As a trio, agree on three AI rules you genuinely believe are smart and fair. Write them down.
2
Take a situation · 2 min
Choose one everyday situation from the next exhibit. Your district on an ordinary day.
3
Play it out · 10 min
Run it through each role, then mark each rule: held, ignored, or could not tell.
Round 1
Write your three good rules
Do not overthink it. The rules already in your handbook are perfect. Three that come up constantly:
01 Only district-approved AI tools may be used.
02 Never put student or staff personal information into AI.
03 AI may assist as a tutor, but the work must be the person's own.
Round 1 timer
05:00
Round 2 / choose one
Everyday situations
Exhibit 302Teacher
6:50am Tuesday. A 7th-grade ELA teacher has 63 graded essays, student names at the top, and a reteach at 9:00. She photographs the stack for a free AI tool: "list the most common errors."
03Student
The rubric allows AI "as a tutor." Maya pastes the whole assignment and asks for an outline with topic sentences, then writes every paragraph herself. She is not sure if she cheated. Neither is her teacher.
04Staff
3:05pm. A truancy letter has to go home today, in Spanish. The office secretary pastes the student's name, address, and attendance record into the first free translation chatbot Google offers.
05Student
A sophomore opens the district-approved study site. Its brand-new built-in assistant pops up: "Want me to write this for you?" The domain is allowlisted. To the filter, nothing happened.
06Program
The night before a board meeting, an assistant superintendent uploads three years of discipline records, student IDs included, to an AI to "find the trends." District data, district purpose, cleared staff.
Every one of these happens somewhere every week. Pick the one closest to home, and as you play it, watch whether anyone in the room would know.
Round 3
Play it out
Run it like a real Tuesday. Be honest about what each role can actually do with the tools they have today.
S Student: comply, or route around the friction.
T Teacher: notice it, prevent it, or prove it.
A Admin: watch the three rules and mark each.
The finding: when a rule was ignored, who would have known, and how? If the honest answer is "no one," write it down.
Round 3 timer
10:00
Compare notes
Which rule broke first?
01 What did the situation do that the policy never anticipated?
02 Could anyone actually see it happen?
03 Was the rule wrong, or simply impossible to enforce?
Hold on to your answer to 02. The rest of this session is about that question.
What you just watched
Your rules were not bad. They were undefended.
Exhibit 4Failure one · Invisible
Most use is never seen at all: personal accounts, phones, and AI that appears inside sites you already allow.
Failure two · Seen, not stopped
What you do see, nothing acts on in the moment. The policy is a request, and the prompt has already been sent.
Failure three · Carried by people
Where enforcement does exist, it runs on human vigilance: teachers watching screens and staff clearing alert backlogs no one can keep up with.
Three failures. The rest of this session takes them one at a time.
Failure one · Invisible
Four ways districts try to enforce
Exhibit 5
Before, at the door
Web filters
Block known sites. Miss embedded AI, new tools, phones, and what gets typed.
During, if watched
In-class monitoring
A teacher watches screens. Blind after the bell, at home, and inside the prompt itself.
After the fact
Assignment review
Inspect the final product. Catches the essay, not the process or the data already shared.
Every prompt, everywhere
Point-of-use governance
Sees, enforces, and protects as it happens, on device. Where Tenet works.
Each earlier method guards one slice of time, and each can only defend what it can see. Only point-of-use governance covers the moment the prompt is sent.
Failure one · Why each defense misses
Enforcement is for sale. The currency is your people.
Exhibit 6 / drag the dialTwo axes decide whether a rule survives a normal day: can you see it, and can you act on it. Human effort buys movement, at a price, with diminishing returns, and only so far.
Human effort invested in enforcement
The binder on the shelf0%
At zero effort, only point-of-use governance is already in the target quadrant.
Enforcement at point of use →
Real governance
All intent,
no control
no control
Written policyhumans are the mechanism
Web filterflag backlog grows too
Point-of-use governancenot connected to this dial
Visibility into actual use →
Failure one · The evidence
The filter was on. The gaps were still there.
Exhibit 7Blocklists chase domains, but AI now spawns inside approved sites, on personal phones, and behind home wifi, exactly like scenario 05. A district already running an AI-governance web filter surveyed its high school teachers. With the filter live:
92%
of teachers do not trust students to use AI appropriately without direct human supervision
Rated 1 or 2 of 5. 70% chose the lowest rating. Not one teacher strongly agreed.
70%
say AI misuse is already a problem in their own classroom.
78%
say it is a problem at home, where the filter does not reach. 61% strongly agree.
The takeaway
The answer is not to block harder. Lock everything down and students simply leave for ungoverned devices. The goal is to make the governed path the easy path.
District baseline teacher survey, grades 9 to 12, 2026. AI-governance web filter and transparency tooling already in place.
Failure two · Seen, not stopped
Guess the number
Exhibit 8Suppose you fix visibility. One school we worked with switched on a monitoring tool that flags risky chats. In its first two days, across more than 1,500 student conversations, how many flags did it raise?
?flags in 2 days
across 0+ conversations
On paper, this is maximum visibility, a flag every thirty seconds. But with no realistic resources to act on them, it is as good as tracking nothing at all. Flagging told them it happened. Nothing stopped it. Visibility you cannot act on is not safety. It is noise.
TransparencyFlaggingActive governance
Transparency shows you the logs. Flagging sends the alert. Both arrive after the fact. Only active governance acts while the prompt is still on the screen.
Failure three · Carried by people
Control is manual, and it is wearing teachers down
Exhibit 9Who feels in control
43%
of teachers feel in control of AI use in their classroom.
Fewer than half. And the ones who do are not relaxed about it.
What that control runs on
90%
of those in-control teachers still do not trust students to use AI unsupervised.
70% rate that trust at the very lowest, and only 10% trust them. 30% say monitoring AI already eats meaningful class time. Their control is vigilance, not the tool.
The human cost
The teachers who feel in control are the ones watching every session. The rest may simply not see what they are missing. Either way, a policy that runs on teacher vigilance is borrowing against the people you can least afford to burn out. Governance should carry that weight, not the teacher.
District baseline teacher survey, grades 9 to 12, 2026.
The second arena, revisited
The same three failures, inside your own walls
Exhibit 10An internal AI program is not a vendor list and a training day. The same three failures appear the moment staff start working, and staff prompts carry record-level data.
Define the approved lane, then defend it
An allowlist no one enforces is the Level 1 policy again, just for adults.
Protect the data at the moment of paste
Rosters, IEPs, and personnel files leave through staff prompts, not student ones.
Govern what you build, not only what you buy
An internal assistant on district data needs the same guardrails you ask of any vendor.
The distinction underneath all three failures
Policy is not governance
Exhibit 11| Policy | Governance | |
|---|---|---|
| Form | A document | A system |
| Sees actual use | No | At the point of use |
| Tested | Rarely, until an audit | Every prompt, in real time |
| Who carries the weight | Teachers and staff | The system itself |
| Covers staff and internal use | Seldom addressed | By design |
Governance sees what policy cannot, acts where flagging stops, and carries what vigilance is dropping. We keep purchasing the first and reporting it as the second. Source: Tenet, 2026.
In one line
A policy without enforcement
is a suggestion.
Where this leads
The gap we built Tenet to close
Governance inside the AI conversation, on the device, in real time, for students and staff. It sees the invisible, acts in the moment, and carries the weight so your people do not.
Prompt
A student or staff member types or uploads.
→
Tenet, on device
Redacts personal data, applies policy, checks safety, before anything leaves.
→
AI tool
Receives only what is allowed. Response rewritten on screen.
Sees the prompt
Redacts names and personal data before anything reaches the AI, typed or uploaded.
Enforces the rule
District, classroom, and staff policy applied live across every major AI tool.
Catches the crisis
On-device safety detection surfaces self-harm and bullying as it appears.
Selected evidence
What the research shows
Sources54% of U.S. teens have used AI chatbots for help with schoolwork; one in ten do all or most of their schoolwork with one.
Pew Research Center, 20267 in 10 teens have used a generative AI tool; 40% used it for school assignments, and 46% of those did so without the teacher's permission.
Common Sense Media, 2024About 6 in 10 teens say their school has no rules for generative AI, or they are unsure whether it does.
Common Sense Media, 202484% of high school students use generative AI for schoolwork, up from 79% just four months earlier; 69% use ChatGPT for assignments.
College Board, 2025With student-monitoring software, 78% of teachers said students were flagged for discipline and 44% said a student was contacted by law enforcement from an alert.
Center for Democracy & Technology, 2022OpenAI retired its own AI-text detector over a "low rate of accuracy"; such detectors also disproportionately misflag non-native English writers.
OpenAI / TechCrunch, 2023Thank you
Governance is not what you
wrote. It is what you enforce.
Caleb Vail Founder, Tenetcaleb@truemadeai.comtruemadeai.com773.766.5694
Next step
Scan to register for the implementation webinar