K-12 AI Tools Dataset: a School Verdict for Every AI Tool Domain
Each of 11,853 AI tools carries a decision written for schools: block, allow with controls or allow. The verdict, its reason and 13 risk factors sit next to the domain categories you already use, so an edtech or device platform can apply a district-ready AI policy in one lookup.
Who this dataset is built for
Products that serve schools now have to answer one question for every AI site a student opens: is it allowed here? This dataset answers it in advance, with a reason a district can read.
Edtech and classroom platforms
Show teachers which AI tools are approved for class and which need a data agreement first. Link out only to tools that match the district's rules.
Student-device platforms
Push the same policy to Chromebooks, iPads and Android tablets. Domain lists and app IDs come from one source.
Student safety vendors
Add a school verdict and a reason to every AI domain in your alerts and reports. Companion chatbots and deepfake tools are flagged before a student reaches them.
Researchers and analysts
Measure how AI tools handle user data across 18 categories and 165 subcategories. Every terms field links back to the vendor's own wording.
Why schools need a decision, not just a list
Blocking every AI tool removes tutoring, translation and read-aloud support that many pupils depend on. Allowing every AI tool opens the door to essay writers, companion chatbots and face swap apps.
The first question is about training
Parents want to know whether a tool learns from what their child types. As of October 2026, 671 AI tools in the data state that they train on user data, and 215 more do so by default unless the user opts out.
Written AI policies are becoming law
Several states now require districts to adopt an AI policy in writing. Florida's new rules go further and require parental consent before students use AI.
Reviewers ask why
A blocklist only shows what is blocked. Boards, auditors and families also want the reason, and every verdict here comes with one.
Five parts, one domain key
The data describes each AI tool and how it treats user data. The rules turn that description into a school decision, and the policy kit turns the decisions into documents a board can adopt.
AI tools database
11,853 AI tools in 18 categories and 165 subcategories. Rebuilt every day.
K-12 policy rules
A verdict and its reason for every tool. Derived from 18 category rules and 165 subcategory rules.
13 risk factors
Data sovereignty, abusive purpose and training on user data. Plus 10 factors read from vendor terms.
AI apps for MDM
2,101 AI apps for iPad and Android. Each with its app ID and its AI tool.
K-12 AI Policy Kit
Policy, parent letter, staff and student guidelines and a review checklist. All built on the same rules.
Delivered next to your categories: the AI tools are a slice of our 120 million domain database, so the K-12 fields join on the same domain column. A single record can carry the web filtering category, the IAB path and, for AI tools, the school verdict.
Part 1: the AI tools database underneath
Everything starts from the full domain database of about 120 million categorized domains. AI tools are picked out of it, placed in a category tree of their own and rebuilt every day.
18 categories, 165 subcategories
The tree runs from AI tutors and captioning to companion chat and face swap. Schools can allow types of AI instead of choosing between all and nothing.
Daily rebuild
About 300,000 newly registered domains are reviewed every day. New AI tools join the list and receive a verdict in the next build.
Every enforcement point
Web filters, firewalls, DNS filters, Chromebooks and MDM all read from the same source. A tool blocked on the network is blocked on the tablet too.
Delivery formats
Files are delivered for data teams and for the systems that enforce the policy.
Part 2: K-12 policy rules and the verdict split
Every AI tool receives exactly one of three verdicts. The split below is from the build of 17 October 2026 and changes slightly as new tools arrive.
Block
Blocked on school networks and devices. Not used for schoolwork on any device.
Allow with controls
Used only with a school account, a signed student data agreement and a teacher's approval for the class.
Allow
Low-risk tools that do not generate open-ended text, such as mind-mapping. Safety tools such as content moderation also land here.
From tool to verdict in five steps
The same five steps run for every tool, so two similar tools always end up with the same answer. A district that disagrees with an outcome can see exactly which step produced it.
Hard blocks first
A tool with an abusive purpose, a high risk rating or high data-sovereignty exposure is blocked. Nothing later in the process can undo this.
Baseline from the labels
Each subcategory label brings its K-12 rule, or its all-sector rule where no K-12 rule exists, and the strictest one wins. A tool with no ruled label takes its category default.
Tightening for data use and risk
Training on user data by default makes the verdict two steps stricter. Training unless users opt out, medium risk and medium sovereignty exposure each add one step.
Relief for AI-enabled products
An established product that has added AI features is controlled instead of removed. Hard blocks from step 1 never get this relief.
Relief for education tiers
When training on data was the only reason for a block and the vendor's education tier does not train, the tool becomes allow with controls on that tier.
Example rows from the verdict file
These rows are shown as they appear in the data, with the reason field unchanged.
| AI tool | Verdict | Reason | Trains on user data |
|---|---|---|---|
| ChatGPT | Block | General assistants & chatbots (K-12 rule) | Trains unless opted out |
| Claude | Block | General assistants & chatbots (K-12 rule) | Trains unless opted out |
| Microsoft Copilot | Allow with controls | General assistants & chatbots (K-12 rule); AI-enabled platform, control rather than block | Trains by default |
| Grammarly | Allow with controls | Grammar, style & proofreading; trains unless opted out; AI-enabled platform, control rather than block | Trains unless opted out |
| illumi | Allow | Mind-mapping & diagramming (K-12 rule) | Does not train |
| Sightengine | Allow | Content moderation | Does not train |
The 18 category rules behind the verdicts
Each AI category has a default for schools and a short rationale that a district can read and change. Subcategory rules then open up what learning and accessibility need inside stricter categories.
| AI category | K-12 default | Why, in short |
|---|---|---|
| Education & Learning | Controls | Made for schools: tutors, quizzes, course building and grading. The vendor has to sign a student data agreement. |
| Code & Development | Controls | Computer science is a K-12 subject. Code assistants are now part of how students learn to write software. |
| Productivity & Collab | Controls | Documents, slides and task boards that staff and older pupils already use. The school account keeps them in check. |
| Design, 3D & Creative | Controls | Design, CAD, 3D and game making are taught in class. The teacher picks which tools a class may use. |
| Robotics & Embodied AI | Controls | Robot kits and computer vision show up in STEM lessons. Low risk, clear teaching value. |
| Security & Detection | Controls | Security and fraud tools belong to staff. Detection of AI content, deepfakes and harmful posts is allowed outright. |
| Text & Language | Block | Open text generators can write the whole assignment. Open chat also shows minors unfiltered output and collects personal data at sign-up. |
| Image & Visual | Block | Open image generation is the easiest way to make explicit or hurtful pictures of real people, classmates included. |
| Video | Block | Text-to-video, avatars and dubbing create fake media of real people. The risk matches image tools and looks more real. |
| Audio, Voice & Music | Block | Voice cloning and voice changers make impersonation and bullying easy. Speech-to-text and text-to-speech are opened up as assistive tools. |
| Lifestyle & Entertainment | Block | Companion chat, dating, astrology and games carry the highest risk for minors. Emotional dependence, sexual content and data collection all meet here. |
| Search, Knowledge & Docs | Block | Answer engines hand over a finished answer instead of sources. Chat-with-PDF tools summarise the set reading for the student. |
| Models & Infrastructure | Block | Raw AI APIs and hosting platforms skip the safety filters that consumer apps add. A student with an API key reaches unfiltered output. |
| Agents & Automation | Block | Agents act on accounts, browse and send data with nobody checking each step. |
| Data, Analytics & Research | Block | Business analytics and scraping tools have no pupil use. They often take in files that may contain student records. |
| Business & Vertical | Block | Software for finance, HR, legal and healthcare professionals. None of it is built for pupils. |
| Marketing, Sales & SEO | Block | No classroom purpose. These tools gather contact data and send messages in bulk. |
| Customer Support | Block | Helpdesk bots and call-centre agents are business software. |
165 subcategory rules: where learning and access are protected
Of the 165 subcategories, 112 are blocked, 48 are allowed with controls and 5 are allowed. The selection below shows how assistive and learning tools stay open inside categories that are blocked by default.
| Subcategory | Rule | Rationale |
|---|---|---|
| Speech-to-text & transcription | Controls | Assistive technology for pupils with dyslexia or motor impairments. |
| Text-to-speech | Controls | Reads text aloud for pupils who struggle with reading. |
| Subtitles & captions | Controls | Access for deaf and hard-of-hearing pupils. |
| Translation & localization | Controls | Helps pupils who learn in a second language. Controls keep personal data out of what they type. |
| AI tutors & personalized learning | Controls | Core classroom use, with a data agreement and age-appropriate design required. |
| Mind-mapping & diagramming | Allow | Organises the student's own ideas, with little data or content risk. |
| Writing & long-form | Block | Writes complete essays, the central academic integrity risk. |
| Paraphrasing | Block | Rewords copied text so plagiarism checks miss it. |
| AI companions & character chat | Block | Emotional dependence and sexual role-play with minors. Blocked in every sector. |
| Face swap & effects | Block | A direct route to deepfakes. Blocked in every sector. |
Part 3: 13 risk factors on every AI tool
Three core risk axes describe what a tool is and where its data goes. Ten more factors come from each vendor's published terms, and every one carries the exact clause and a link to the vendor's terms page.
Data sovereignty
Where data is stored and processed, and which jurisdiction applies.
Abusive purpose
As of October 2026, 296 tools are flagged as NSFW, deepfake, uncensored or nudify tools. Each one is a hard block.
Trains on your data
Whether the consumer tier uses what people type to train its AI.
Opt-out available
Can users stop training on their data?
Opt-out scope
Does the opt-out cover all data or only future data?
Enterprise no-training
Do paid or education tiers exclude customer data?
API no-training
Is API traffic left out of training?
Feedback exception
Are ratings and thumbs up or down still used for training?
Retention
How long are prompts kept for monitoring?
De-identified use
May "anonymised" data still be used?
Human review
Can vendor staff read conversations?
Zero-retention option
Is there a mode that stores nothing?
Subprocessor training
May third parties train on the data?
Training on user data across all 11,853 tools
Counts as of October 2026, checked against each vendor's terms.
Silence is a signal too: under the K-12 rules a tool that does not say whether it trains cannot get a plain allow. At most it reaches allow with controls.
One delivered record
Each field below also carries the vendor's clause and a link to its terms in the full file.
trains_consumer_default yes
optout_available yes
optout_scope future_only
feedback_exception yes
deidentified_use permitted
human_review yes
enterprise_no_training yes
api_no_training yes
k12_verdict block (General assistants & chatbots, K-12 rule)
Part 4: AI apps for MDM
A web filter does not see an AI tool that a pupil opens as an app. The app list carries the same tools and categories, so the policy on the network also holds on school tablets.
AI apps
Each linked to its AI tool and category, as of October 2026.
Android apps
With package names.
iOS and iPadOS apps
With bundle IDs.
New apps
App stores are checked every week for new AI apps.
Import files included
Microsoft Intune
Restricted app lists for iOS and for Android.
Jamf
A configuration profile ready to upload.
Android Management API
A blocked apps policy.
Plain ID lists
Bundle ID and package name lists for any other MDM.
18 category files
One app list per AI category, for policies that open whole groups.
Tool link
Every app points to its AI tool domain and verdict.
Part 5: the K-12 AI Policy Kit
The kit turns the verdicts into the documents a school board adopts. It uses the same three verdicts and the same five steps, so what the board approves is what the filter enforces.
AI Acceptable Use Policy
Ready for a board vote, in 14 sections with 2 annexes.
- Purpose, scope and definitions
- The three verdicts
- How tools are sorted into categories, with a rating guide and examples
- What allow with controls requires
- Student data and privacy (FERPA, COPPA, CIPA)
- Academic integrity and staff use
- Requesting a new tool, and exceptions
- Roles, violations and policy review
- Annex A: K-12 category rules
- Annex B: Approved AI Tools List
Letter to parents
One page in plain words.
- The three decisions, with examples
- How the child's data is protected
- Consent for students under 13
- AI in schoolwork
- How parents can help at home
Staff guidance
For teachers and staff, in 6 sections.
- What you can use
- Using an approved tool
- Never paste into an AI tool
- Requesting a new tool
- Handling suspected AI misuse fairly
- If something goes wrong
Student guidelines
Two versions for two age groups.
- Grades K-5: 8 short rules with a signature line
- Grades 6-12: 9 guidelines on approved tools, AI levels per assignment, disclosure, privacy, respect, fact checking, wellbeing and reporting
Tool review checklist
One page for IT and curriculum staff.
- Describe the tool
- Hard blocks
- Baseline rule
- Tightening for data use and risk
- Relief, ending in a verdict
Rules and ready-to-import files
The rules as data and the verdicts as filter files.
- Category rules, subcategory rules and decision layers
- Google Admin blocked URLs
- Chrome policy JSON
- DNS RPZ zone and hosts file
- Block, allow and controls domain lists
In the kit's own words
Short passages quoted from the kit documents. Bracketed text is filled in by the district.
It has three goals, in this order:
- Protect students, who are minors.
- Protect academic integrity.
- Keep vetted, useful tools available for teaching and learning.
An AI tool that cannot be placed in any category is blocked until it is reviewed. A tool with the verdict allow with controls may be used only after it has been added to the Approved AI Tools List.
People make decisions about students. Grades, placement, discipline, special education and other decisions about students are made by staff, not by AI.
Speech-to-text, text-to-speech, captioning and translation tools are already allowed with controls.
Tip: remove identifiers first, for example "Student A" instead of a name.
AI detectors can be wrong, so a detector result is never treated as proof on its own: teachers always talk with the student first.
Check what AI tells you. AI can be wrong, can be biased and can make up sources.
I check with a grown-up. AI can be wrong. I check what it says with my teacher or a book.
AI-enabled product (not a hard block from Step 2): allow with controls.
Reviews go faster when the vendor has an education tier that does not train on user data, signs the data-privacy agreement, says where data is stored and supports district sign-in.
What states expect from schools
A summary of state and federal rules on AI in K-12 schools, as reported in October 2026. Check the current text with counsel before relying on it.
Consent and an off switch
The State Board of Education approved AI rules in September 2026. Parents must give consent before a student uses an AI product, and a non-AI alternative must be offered.
AI that simulates an emotional relationship with a student is barred, and staff must be able to turn AI off at once. District policies are due by 1 July 2027.
A policy in every school
Every district, community school and STEM school had to adopt an AI policy by 1 July 2026. The state published a template policy covering ethics, academic integrity, data privacy and third-party tool review.
Keep it and report it
Local boards and public charter schools have to keep a policy on AI use. They also have to report it.
Written before 2027-28
Each district needs a written AI policy before the 2027-28 school year. It lists approved and prohibited uses and protects student data.
Parents can opt their child out of student-facing AI tools.
Companion chatbots
Both states have passed laws on AI companion chatbots and children. California's was signed in September 2026, and Colorado's applies from 1 January 2027.
Congress and COPPA
Bills on AI companions for minors are before Congress, and the amended COPPA rule applies from April 2026. AI companions already carry a block verdict in this dataset.
For platforms: one dataset supports districts in every one of these states. Each district can tune a rule and keep its own reasons on record.
How data teams put it to work
Most buyers already join domain categories into their product. The K-12 fields ride along on the same key.
Join on the domain
Match the AI tool file to the domains you already categorize. Every AI tool gains its verdict, reason, category, subcategories and risk factors.
Pick a rule set
Use the K-12 defaults as delivered, or move individual category and subcategory rules for a district. The decision layers show which tools change.
Push to every device
Send domain lists to the filter, Chrome policy to Chromebooks and app lists to Intune, Jamf or Android Management.
Show the reason
Display the reason field when a teacher or parent asks why a tool is blocked. Hand the policy kit to districts that need a board document.
K-12 AI tools dataset: FAQ
What is in the K-12 AI tools dataset?
Every AI tool domain carries a K-12 verdict (block, allow with controls or allow), the reason for it, its category and subcategories, and 13 risk factors. AI app lists for MDM and the K-12 AI Policy Kit come with it.
How does it fit with the domain categories we already license?
The AI tools are a subset of the same 120 million domain database, so the K-12 fields join on the domain you already use. One lookup returns the web category and, for AI tools, the school verdict and risk factors.
Can a platform or district change the rules?
Yes. All 18 category rules and 165 subcategory rules are delivered as data with their rationale, so you can move any rule and recompute which tools change.
Does every AI tool end up blocked?
No. Tutoring, accessibility, coding and translation tools are allowed with controls, and a small group of low-risk tools is allowed. Open chatbots, companion apps, deepfake tools and essay writers are blocked.
How often does the data change?
The AI tool list and the K-12 verdicts are rebuilt every day. App stores are checked every week for new AI apps, and vendor terms are checked again when vendors change them.
Which systems can enforce the verdicts?
Web filters and firewalls through domain lists, Chromebooks through Google Admin blocked URLs and Chrome policy, DNS filters through RPZ and hosts files. School tablets are covered through Intune, Jamf and the Android Management API.
Give every school you serve a ready AI policy
Licences for edtech platforms, student-device and safety vendors, researchers and districts. OEM rights are available for products that serve many schools.
Contact us See pricingProduct names in the examples are trademarks of their respective owners and appear only to identify the tools described. IAB is a trademark of the Interactive Advertising Bureau.