HNK Institute publishes this parent briefing on hnk.org because families are often sold fluency as if it were literacy. Student AI tools can draft, summarize, and suggest. They cannot own a student’s judgment, cannot certify that a claim is true, and cannot accept responsibility when a homework submission misleads a teacher. Human knowledge for new technology means the adult conversation starts with limits, not with feature lists.
This page is a briefing script you can read at a kitchen table. It is not a product review and not a panic essay. Responsible AI education treats limits as content students practice, the same way STEM knowledge practice treats units and definitions as content. If a program never rehearses what tools cannot do, it is running gadget theater.
Limit one: tools do not understand on the student’s behalf
A model can produce a polished paragraph about a science idea while the student still cannot define the key term with the laptop closed. Understanding is what remains when the account is logged out. Parents can test this without becoming engineers: close the interface and ask the student to teach the idea back in ordinary speech. If the voice collapses, the knowledge was on loan.
HNK asks students to write a human-language account first, then decide whether any tool is needed. Opening a tool first and writing the explanation afterward is treated as a literacy gap, not as efficiency. The next assignment is a clearer human sentence, not a more impressive prompt.
Limit two: fluency is not evidence
Fluent answers look finished. Finished is a visual effect. Evidence is a source note that separates a claim from a guess, a limitation note beside any example, and a revision log that shows what changed before anyone showcases the work. Screenshots of chat threads without a human account of what they show are decoration.
Families evaluating programs should ask what artifacts are kept: explanation cards, maps of what is unknown, teach-backs. If the portfolio is mostly exports, the program is collecting machine text. STEM knowledge practice requires artifacts a classmate could mark.
Limit three: tools cannot hold privacy for you
Student AI tools do not automatically protect family data, classmate names, medical details, or school credentials. Pasting private records into a public or semi-public model is a human decision. Literacy includes practiced refusal: what must not be sent, what must not be pasted, what must not be claimed as personal knowledge when it was generated.
A parent briefing should name three refusals out loud. Example: we do not paste other students’ work; we do not upload ID photos to random tools; we do not submit generated text as if it were the student’s unaided voice without disclosure rules the school accepts. Small refusals are how responsible AI education becomes habit.
Limit four: tools cannot replace academic honesty design
Honesty is assignment design, not a warning poster. If an assignment can be completed entirely by pasting the prompt into a model, the assignment is inviting deception. Teachers and parents should ask whether the task requires a human decision the student must still own: a definition, a limitation, a comparison, a teach-back, a lab note tied to something the student actually did.
HNK mentors may not complete the student’s explanation. They may not replace a weak student sentence with a better generated one. They may ask what would make the explanation false. That question protects students from fluent machines and from helpful adults who accidentally erase authorship.
Limit five: tools cannot choose the student’s values
Recommendation engines and chat assistants can suggest next steps. They cannot decide whether a use is fair to classmates, respectful of sources, or appropriate for age. Values questions stay with the human. A parent briefing that only discusses speed and grades will miss the point of AI literacy for students.
Ask the student to list one thing the tool cannot do even when the output looks confident. Keep that sentence beside the device. Confidence without ownership is the common failure mode this briefing exists to interrupt.
How to run a twenty-minute parent briefing
Minute one to five: close the laptop and hear a teach-back. Minute six to ten: read a limitation note the student wrote. Minute eleven to fifteen: name two refusals the family will keep. Minute sixteen to twenty: agree what the school needs to know about disclosure. No need for a vendor demo. If a program requires a demo before it can explain limits, treat that as a signal.
Schools without gadget theater write language targets before device lists. Families can ask for those targets. HNK’s curriculum is built as tracks of explanation-first practice, not as a tour of interfaces that will be replaced next term.
Language to keep and language to drop
Keep verbs of explaining, limiting, citing, and refusing. Drop slogans that cannot be marked: “AI native,” “future proof,” “guaranteed advantage.” If a sentence cannot survive the closed-interface test, it does not belong in a parent briefing as a promise.
When a student hides uncertainty so the work looks finished, name that as a gap. The assignment is not to sound like a model. The assignment is to sound like a student who knows where the map ends.
What this briefing refuses
This page refuses to invent customers, prices, awards, or testimonials. It refuses to treat a generated paragraph as a student’s mind. It refuses to scare families into buying tools or into banning thinking aids without practice. The practical next step is small: write the human account, list one cannot-do, keep a source note, and only then decide whether a tool earns a place in the folder.
If you only remember one line from the briefing, remember this: student AI tools can help with drafting; they cannot do the student’s understanding, honesty, or responsibility. HNK exists to keep that line audible when the interface is fluent enough to make adults forget it.