Skip to main content
Michigan Virtual

The AI Safety Pack

AI can be a useful tool, but using it well means knowing what to look out for. This guide gives students, families, and teachers practical questions and shared expectations for using AI safely and responsibly.

Purpose

This pack helps you make choices about AI that protect learning, personal information, wellbeing, and other people. It gives students, families, and teachers a common way to talk about AI tools, set expectations, and respond when something goes wrong.

This pack is a reference, not legal advice, and should not be a substitute for your school district's own AI guidance or policy. Ideally, you can use this pack to work through key questions for aligning on how to use AI together.

How to use the pack

Start with the section for you. Students can use their section independently to reflect on their approach to AI. Families can adapt the conversation starters for any age. Teachers can use the questions when planning a lesson or reviewing classroom use of AI.

Every section asks the same five questions, so that they can be used cohesively:

  • Does AI help with learning?

  • What happens to our data?

  • Is the AI tool safe?

  • Can I trust what AI says?

  • Is the AI straightforward?

Shared Ideas is provided to help you build a common understanding about AI safety topics, so jump here if you aren’t clear on any terms included in your section. Use the shared safety plan to agree on boundaries for AI use at home and at school, and return to the questions whenever a tool changes or a new concern arises. You can answer "I don't know yet" and decide what to check next, so you do not need technical expertise to start building a shared safety plan.

Getting Help provides links to independent reviews, agency guidance, and further reading. Bracketed numbers point to the corresponding sources throughout the pack, so you can easily find where to learn more.


1 Note that the student section is designed to be completed independently, so it is most suitable for middle and high school students.

For Students

AI can support your learning, but it can't learn for you.

Use these questions before you pick an AI tool and while you use it. A tool that helps with one task can be the wrong choice for another, so it is important to use these questions to guide the decisions you make about AI. You can always ask a human for help or complete your schoolwork in a different way.

Does AI help with learning?

Think of AI as a spotter, not a stand-in. 

Used well, AI is like a spotter at the gym: it can give you a hint, check your form, or suggest a next step. Used poorly, it does the push-ups for you, and you stay exactly as strong as you started. Finishing the work is not the same as understanding it. [1]

Ask yourself: What am I trying to learn? Is AI coaching me or handing me the answer? Could I explain or use this idea on my own afterward?

Try this:

  • Ask for a hint or feedback on an attempt, not the finished answer.

  • Tell the chatbot you don't want it to solve the problem or do the work for you.

  • Make the first draft yours, then ask AI to push on it.

Run the WISE check

Before you use AI on something that matters, consider these four questions. Adapted from Dr. Sabba Quidwai’s WISE framework in Designing Schools [2]:

  • W — Wellbeing: How does this affect me and the people around me?

  • I — Integrity: Am I being honest and fair about how I'm using it?

  • S — Skills: What will I learn by using it, and what might I miss learning?

  • E — Engage: Does this pull me into the work, or let me check out of it?

What happens to our data?

If you wouldn't tell it to a stranger, don't type it into AI.

Every prompt is a trade: you get an answer, and the company gets your data. What companies do with your data varies, some use it to sell you things, others use it for research or model training. Also, some tools store what you type, and turning off model training does not always mean your conversations aren't kept. [3]

Ask yourself: Am I sharing names, photos, voices, passwords, private messages, or details about myself or someone else? What happens to it next?

Try this:

  • Leave out personal details you don't need to share, including voice recordings, video, photos.

  • Swap specifics for generalities: instead of "I'm Sam from Lincoln High and my meds aren't working," try "I'm a high school student — what are some general ways to manage stress?"

  • For schoolwork, use the approved school accounts and tools.

  • Ask before uploading anything that could identify you or another student.

Is the AI tool safe?

Safe isn't one-size-fits-all. It depends on the tool, your age, and how it affects you. A tool that feels safe to one person may not feel the same way to you, which is OK. 

Ask yourself: Is this tool meant for my age and allowed for me to use? Does it pressure me to keep going, ask for secrets, or affect my sleep, friendships, or mood (positively or negatively)?

Try this:

  • Stop using the tool if an interaction turns frightening, sexual, threatening, or pushes you toward harm; and tell someone you trust.

  • Don't make or forward deceptive, fake, humiliating, or sexual images of anyone.

  • Verify urgent requests for money or information through a separate contact you already trust. You may also want to have a secret password that only you and trusted friends and family know. [4,5]

AI companions imitate a personal relationship, but they are not people. For that reason, anyone under 18 is advised not to use social AI companions, and a chatbot should never replace human support or professional care. [4]

Can I trust what AI says?

AI is trained on text, which is not necessarily true, so it is critical that you be the human in the loop.

Ask yourself: Can I find this answer somewhere else? What could be wrong, missing, or unfair? Is it just agreeing with me?

Try this:

  • Open the original source and check that it actually supports the claim. AI can fabricate a source that looks real.

  • Be cautious with answers that only echo your own view or lean on a stereotype. AI can sound confident, but be wrong. 

  • Ask a knowledgeable person about important decisions, including an AI-assisted grade or feedback you think is wrong.

Is the AI straightforward?

You should always know when AI is being used, and what the rules for using it are.

Ask yourself: Does the tool say it's AI and explain its limits? Do I know the assignment rules and who to ask about a concern?

Try this:

  • Clarify the rules before you start using a new AI tool or are using it for the first time for a new type of task.

  • When you need to share how you used AI, say what AI did and what you checked: "I used AI for ___, checked ___, and did ___ myself."

If something goes wrong

Stop using AI immediately and get human help. Call 911 for immediate danger; call or text 988 for emotional or suicidal crisis support. For Michigan school safety concerns, contact OK2SAY at 855-565-2729 or text 652729. See Getting Help for reporting and image-abuse support. [8–10]

For Families

The most important thing you can do as a parent or guardian is to reassure your student that whatever happens with AI, they can come to you.

You can use the questions below to help your student make sense of using AI wherever they come across, including in search, games, school apps, or toys. Your student is likely using AI more than you think, and the goal isn't to lock it down, but rather to actively learn about it together. 

The questions are age-agnostic, so adjust the wording as you see fit. With younger students, it is more important for you to make access decisions yourself and stay involved in any use.

Does AI help with learning?

The best thing you can do is to notice when AI is doing the thinking that was supposed to be your student's and talk to them about the downsides of this.

Start a conversation: "What do you want this to help you learn?" "Is it giving you a hint, or doing the work?"

It may be helpful to use the gym metaphor, because students get it: AI as a spotter helps you get stronger; AI doing your push-ups leaves you exactly as strong as you started. Look together at one AI tool and ask your student to explain what they understand now and how the tool helps them understand more or prevents them from doing so. Ask the school what benefit an assigned tool is meant to provide and what evidence supports it. Keep needed learning and accessibility supports in place. [1]

What happens to our data?

Personal and school accounts don't always protect data the same way. As a general rule, school accounts have more data protection than personal accounts. 

Start a conversation: "What does this app ask you to share?" "Could it identify you or someone else?"

Check what data is collected, why it is needed, how long it is kept, and whether it is shared, sold, used for ads, or used to train AI. Find out whether you can review, export, correct, or delete it. Ask the school about assigned tools, since personal and school accounts may carry different protections. [3]

Is the AI tool safe?

Filters and age labels alone do not establish safety, so it is important to learn more about the tool with your student.

Start a conversation: "Has it ever said something that made you uncomfortable?" "How do you feel after using it?" "What would make it easier to tell me if something went wrong?"

Check age rules and content protections. Agree on boundaries for companion features, private information, purchases, and time spent. Notice whether AI is displacing sleep or relationships. If your student feels attached to an AI tool, ask what the interaction provides and help them find human support. [4]

It is recommended that anyone under 18 should not use social AI companions. It recommends avoiding AI companion toys for children five and under and extreme caution for ages 6–12. [4]

If your student reports harm, begin with "Thank you for telling me" and help without blame. Don't promise complete secrecy or that every harmful image can be removed until you report it and learn more.

Can I trust what AI says?

"How did you check this?" is a better question than "Did you use AI?"

Start a conversation: "How could we check that?" "Does the source say what the AI claims?"

Model opening sources and admitting uncertainty. Ask whose perspective might be missing. For schoolwork, ask how AI-generated feedback is checked and how your student can request human review when they have questions about something.

Is the AI straightforward?

You're entitled to plain-language answers about the tools your student uses.

Start a conversation: "Do we know when AI is being used?" "Are the rules clear?" "Who could we ask about this?" “What evidence is there that it is beneficial?”

Ask for plain-language information about school tools, limits, alternatives, and changes. Share relevant observations from home. Use the shared plan to record boundaries and a contact route, and revisit it together when circumstances change. [6]

Try this: AI disclosure at home

Encourage your student to add a short note to their work explaining how they used AI. For example, "I used AI to brainstorm three title ideas, then wrote the essay myself." It lowers the stress of wondering whether they are allowed to do something and builds trust with you and their teachers by building a habit of being honest about their AI use.

If something goes wrong

Stop the interaction and get human help. Call 911 for immediate danger; call or text 988 for emotional or suicidal crisis support. For Michigan school safety concerns, contact OK2SAY at 855-565-2729 or text 652729. See Getting Help for reporting and image-abuse support. [8–10]

For Teachers

Treat AI like an overconfident intern: fast and eager, doesn’t have context or background knowledge, and is sometimes confidently wrong. Similar to supervising an intern, you must be involved in directing and reviewing their work, particularly when it directly affects students. 

Use the five questions to decide whether an AI use is aligned with the learning goals you have for your students. Follow the school's review process for the specific tool, account, feature, and use. A product marketed for education still needs scrutiny. [6]

Does AI help with learning?

AI can make it look like a student has mastered something that they haven’t. 

Identify the intended learning and the work students need to do themselves, then examine evidence relevant to the task and your learners. [1]

Before the task, complete: "Students are learning ___. AI may assist with ___. Students will demonstrate ___." Plan an accessible alternative and preserve needed accommodations.

What happens to our data?

It is important to be careful with data and what you upload for all tools, regardless of whether they have been approved by your school or not. 

Confirm what the approved use allows. Check collection, storage, sharing, training, and deletion terms for the actual account. Use fictitious information in demonstrations and collect only what the task requires. [3]

If terms or permissions are unclear, use the established review process before entering student information, including audio, video, and photos. Don't place private data into a new tool to ask whether sharing it is safe.

Is the AI tool safe?

The final review should always be completed by you. Don't hand students AI output (e.g., worksheets, letters of recommendation, feedback), or ask them to test for dangerous outputs, before you've reviewed and done testing yourself. 

Test AI tools as a student for age suitability, harmful content, and foreseeable misuse. Consider companion features, emotional dependence, scams, and peer harassment, not just classroom content. Give students a way to stop, choose an alternative, and report a concern.

Immediately stop use when a serious concern emerges. Support the affected student and follow the relevant response procedure, even if an assignment rule was also broken. For suspected abuse or neglect, follow Michigan reporting requirements; an internal notification alone is not sufficient. [7]

Can I trust what AI says?

Verify before you trust AI, regardless of how confident it seems. It is critical to keep consequential decisions under your expert judgment.

Check generated materials for accuracy, bias, accessibility, and fit. Review AI-generated grades or feedback before students receive them, and offer an explanation and a way to challenge errors. [6]

When you suspect misuse, lead with a conversation, not an interrogation. "Walk me through how you got here" tells you more than any detector will. It is also helpful to look at a pattern of use rather than one assignment, and treat most cases as a teaching moment. A detector score, a style change, or difficulty answering orally is not proof of cheating. Follow the school's academic integrity process. [11]

Is the AI straightforward?

Model the disclosure habit you want from students.

Before a task, explain the purpose, the allowed assistance, disclosure expectations, data boundaries, alternatives, and the help route. Provide understandable information to families and communicate changes.

Invite feedback about confusing rules, inaccurate responses, and barriers. Review whether the use is helping students, and adjust, pause, or stop it when needed. [6]

If something goes wrong

Stop the interaction and get human help. Call 911 for immediate danger; call or text 988 for emotional or suicidal crisis support. For Michigan school safety concerns, contact OK2SAY at 855-565-2729 or text 652729. See Getting Help for reporting and image-abuse support. [8–10]

Shared Ideas

The Shared Ideas provide students, families, and teachers with a common set of facts. 

How AI actually works (the short version)

Most of the tools in this pack are built on large language models (LLMs). They are not thinking or sentient; they generate a response by predicting a likely next word, over and over, until it reads like a finished answer. That single fact explains most of what feels strange about AI:

  • It can't show its work. There is no chain of reasoning to check behind an answer, just a statistical path the tool can't retrace.

  • It fabricates. Because it is built to keep the response flowing, it will sometimes state false things in a confident tone — a made-up quote, a study that doesn't exist, a citation that looks real. Verify facts, quotes, citations, and numbers every time. [1]

  • It is designed to be agreeable. These tools tend to flatter and agree rather than push back, so "that's a brilliant idea" isn't evidence that it is.

  • It carries bias. It is trained on human-made data, so it reflects human bias. Ask it for a picture of a "doctor" and you'll often see what that means.

Personal information: what to protect

Personally identifiable information (PII) is anything that could pin down who a specific person is. For everyone in the building, the rule is the same: keep PII out of any AI tool the district hasn't approved. That includes:

  • Names: a student's, a staff member's, or a family member's.

  • Locations: school name, home address, the specific places someone spends time.

  • Identifiers: student IDs, dates of birth, record numbers.

  • Records: grades, assessments, IEPs, 504s, health and discipline records.

  • Sensitive details: medical, mental health, family, or financial information.

Many consumer AI tools store what you type and may use it to train future models. Education accounts a district has vetted may be different: they typically operate under a data agreement. When in doubt about a specific tool, ask before you type and use your school account. [3]

Don't skip the struggle

Learning takes effort, and that effort is the key to learning. When a student wrestles with a hard problem and works out why an approach failed, they build understanding that lasts. Handing that thinking to AI finishes the task and skips the part that would have built the skill. [1] It is important to ask: is AI helping me think, or thinking for me?

Shared AI Safety Plan

Download Printable Form

Complete this together for a specific tool or use. Students can contribute experiences and questions; adults can set appropriate protections. Record what you know, what is still unclear, and how you will follow up.

Tool ____________________

What you will use it for ____________________

Review date ____________________

Reviewer(s) ____________________

For each question, write your agreement and any remaining questions.

Does AI help with learning?

Agree on the learning goal and how AI can help with schoolwork. How will you know it helps you learn?

Our agreement: ____________________________________________

What happens to our data?

Agree on what may be shared, the settings to use, and your open questions about how the data will be used.

Our agreement: ____________________________________________

Is the AI tool safe?

Agree on use limits, signs to stop, and where to report concerns. Check the Getting Help section.

Our agreement: ____________________________________________

Can I trust what AI says?

Agree on how AI outputs will be checked and how to request human review of an error or unfair decision, if applicable.

Our agreement: ____________________________________________

Is the AI straightforward?

Agree on where expectations will be shared, how AI use will be disclosed, and how changes will be communicated.

Our agreement: ____________________________________________

Decision: Use within the agreed limits / seek more information / choose another approach

Review again on (date) ____________________

Review sooner if features or terms change, someone reports harm, or the use is not helping learners. Share what changed and why. A completed plan is a record of agreements, not a guarantee of safety. [6]

Getting Help

Stop an unsafe interaction and seek human help. You can ask for support even if you clicked a link, shared information, or broke a rule. If someone tells you about harm, listen, thank them, and explain the next step.

Immediate danger or a crisis

Call 911 for immediate physical danger or a medical emergency. Call or text 988 for emotional distress or a suicidal crisis; you can also seek help for someone you are worried about. Don't wait for a chatbot to respond. [8]

School safety concerns

Use your school's reporting route. Michigan's OK2SAY also accepts confidential school safety tips: call 855-565-2729, text 652729, or visit michigan.gov/ok2say. Call 911 for emergencies rather than waiting for a tip response. [9]

Threats and sexual image abuse

If someone threatens to share sexual images or demands money or more images, do not pay or send more. Stop responding, report and block the account, and seek help. Keep existing threatening messages; don't delete an account just to hide what happened. [10]

Do not request, download, or forward sexual images of minors to collect evidence or show what happened. Note non-explicit details such as usernames, URLs, and times. Report suspected child sexual exploitation, including AI-generated sexual abuse images, at cybertipline.org. [10]

NCMEC's Take It Down may help limit sharing of eligible images already on your device. Don't download an image to use the service. It works with participating platforms and cannot guarantee removal everywhere. Visit takeitdown.ncmec.org for eligibility and instructions. [10]

Scams and impersonation

Pause before sending money, sharing a password, or following an urgent instruction. A familiar voice or image may be fabricated. Contact the person or organization through a number or address you already know is genuine. Report suspected fraud at ReportFraud.ftc.gov. [5]

Private information shared by mistake

Stop further sharing and use the school's privacy or security reporting route, or the service's support route for personal use. Explain what was shared without circulating the sensitive material again. Ask about restricting access, deletion, and any other steps needed. [3]

Suspected child abuse or neglect

Follow Michigan's mandated reporting requirements when they apply. Reporting internally or telling a family alone does not fulfill those duties. Use MDE's Mandated Reporting guidance and its MDHHS reporting links; don't wait for a vendor to investigate. [7]

Local contact ____________________

Alternative contact ____________________

Where to Learn More

Look for the exact product, feature, account type, and review date. A privacy rating does not establish a learning benefit or emotional safety. If a review is missing or old, ask more questions.

Common Sense Media

Use AI risk assessments (https://institute.commonsensemedia.org/risk-assessments) for product and use-case findings, and privacy evaluations (https://privacy.commonsense.org/evaluations) for data-practice reviews. Read the findings behind a rating and check what was actually evaluated.

Internet Safety Labs

App Microscope (https://appmicroscope.org/) provides privacy information for tested mobile apps. As of September 28, 2026, its site warns that displayed scores may be incorrect while score logic is being revised. Check the dated underlying findings and do not rely on those scores until the notice is resolved.

Mozilla Foundation

Mozilla's current privacy review work appears in Nothing Personal (https://www.mozillafoundation.org/en/nothing-personal/privacy-not-included-sunset/). Its Privacy Not Included guide is archived. An older review can raise useful questions but may not describe today's product.

Public Interest Privacy Center

The K–12 Privacy Policy Guide (https://publicinterestprivacy.org/privacy-policy-red-flags/) helps you spot concerns such as broad data-sharing terms or missing deletion procedures. This is an initial screen; school tools still need the school's formal review.

Ask the school

Ask which tools and uses have been reviewed, what restrictions apply, and how to raise an unresolved question. Confirm the account and feature, not just the product name.

Explore further

Sources

1. AI and learning evidence (https://ies.ed.gov/learn/blog/ai-k-12-education-good-bad-and-guardrails-consider) IES review; see also Bastani and colleagues, PNAS (2025), doi.org/10.1073/pnas.2422633122. The high school mathematics study found that tutoring safeguards mitigated a loss in unassisted performance; it did not show an improvement on the unassisted test.

2. Dr. Sabba Quidwai's WISE framework (https://designingschools.org/studentwebinar) Credit: Quidwai, S., Designing Schools: The Future Is a Place We Create. The wellbeing, integrity, skills, and engagement reflection is adapted using Michigan Virtual's Student Guide to AI wording.

3. FTC COPPA guidance (https://www.ftc.gov/business-guidance/resources/complying-coppa-frequently-asked-questions) Section N addresses school uses and data questions. Also see MDE Pupil Privacy at michigan.gov/mde/pupil-privacy and U.S. Department of Education student privacy resources at studentprivacy.ed.gov.

4. Common Sense Media social AI companion assessment (https://institute.commonsensemedia.org/risk-assessments/social-ai-companions) Updated July 2025. Its recommendation against social companions for under-18s is an organizational safety recommendation, not a general legal ban. For younger children, see Common Sense AI toy guidance (January 2026).

5. FTC guidance on fake emergency scams (https://consumer.ftc.gov/articles/scammers-use-fake-emergencies-steal-your-money) Includes voice cloning and verifying urgent requests.

6. MDE artificial intelligence guidance (https://www.michigan.gov/mde/services/academic-standards/educational-technology/artificial-intelligence) Includes the Comprehensive Guide for Districts. UNICEF's Guidance on AI and Children, version 3 (2025), also informs inclusion, participation, and shared safeguards.

7. MDE Mandated Reporting (https://www.michigan.gov/mde/services/ed-serv/educator-conduct/mandated-reporting) Use the current procedures and linked MDHHS information for suspected abuse or neglect.

8. SAMHSA 988 resources (https://www.samhsa.gov/mental-health/988/partner-toolkit) Crisis support information and public-facing help materials. Immediate physical danger or medical emergencies require 911.

9. Michigan OK2SAY (https://www.michigan.gov/ok2say) Official contact and emergency instructions.

10. OJJDP Sextortion Victim Resources (https://ojjdp.ojp.gov/publications/sextortion-victim-resource-sheet.pdf) Use with NCMEC's CyberTipline at cybertipline.org and Take It Down FAQs at takeitdown.ncmec.org/faq/.

11. Liang and colleagues on AI detector bias (https://doi.org/10.1016/j.patter.2023.100779) Patterns (2023). Findings apply to tested tools and samples and support caution about false positives and fairness.

Additional foundations: EDSAFE AI Alliance, SAFE by Design (supplied report); Erin Mote's Senate HELP testimony (June 16, 2026); and Brookings Institution, A New Direction for Students in an AI World (2026). These inform the pack's approach to safety, learning, and human relationships; policy proposals are not treated as adopted Michigan procedures.