California Model AI Policy: Detectors Can't Decide Alone

The California model AI policy for schools does two things at once, and the combination is smarter than either half alone. It says that “submitting AI-generated content without clear disclosure constitutes plagiarism and academic dishonesty.” And it says that “AI detection software should not be used as the sole basis for disciplinary action or grade penalty.” In other words: students must be honest about AI, and schools must be fair about proving it. Published by the California Department of Education under a 2024 state law, the policy is a template districts can adopt, not a mandate. But its approach is a useful map for any family.

Key Takeaways

• The California Department of Education’s model policy, written under Senate Bill 1288 (2024), treats undisclosed AI-generated content as plagiarism and expects students to disclose AI use when they submit work.
• It bars AI detectors from being the sole basis for discipline; concerns must rest on “educator evaluation of student engagement, alternative assessment, and documented evidence of authorship,” with clear ways for families to rebut claims.
• Teachers decide what AI use is allowed on each assignment, and must explain it to students and families.
• The policy is voluntary: the CDE says compliance “is not mandatory,” so your district may have adopted all, part or none of it.
• At home, the two habits that match this policy are disclosing AI use and keeping a record of the work’s process.

What the California model AI policy says

The California model AI policy is a 26-principle template published by the California Department of Education for school districts and charter schools. It was required by Senate Bill 1288 (Becker, 2024) and developed with an Artificial Intelligence Working Group. The CDE is explicit that it is “merely exemplary, and compliance with any information or guidance in this document is not mandatory.” Districts are encouraged to adapt it.

The principles most relevant to students and parents:

Principle	What it says	What it means for your child
1. Plagiarism, Disclosure & Citation	”Submitting AI-generated content without clear disclosure constitutes plagiarism and academic dishonesty”	Hiding AI use is treated like copying
2. AI Detection Software	Detectors “should not be used as the sole basis for disciplinary action or grade penalty”; concerns rely on educator evaluation, alternative assessment and documented authorship, with clear rebuttal criteria	A detector score alone shouldn’t cost your child a grade
5. Instructional Discretion	Teachers decide “when and how AI may be used in individual assignments,” and permitted use “may vary from one assignment to another”	Rules can differ by class and by assignment
6. Modeling & Discussion	Educators “should disclose their own AI use”; students “are likewise expected to disclose their use of AI when submitting work”	Disclosure goes both ways
9. Data Entry Restrictions	No education records or confidential information in AI tools unless authorized	Your child’s records shouldn’t go into chatbots
11. Parent/Guardian Review Rights	Guardians can request “timely human review of any AI-generated outputs that inform decisions about their child”	You can ask for a human to check AI-influenced decisions
18. Grading & Assessment	Students should be told when their work goes to AI for grading; the teacher “retains sole authority to determine final grades”	A human decides the grade

There’s more: AI literacy as a core competency, privacy-by-design rules that “prohibit its use for training public AI models or advertising,” vendor contract safeguards and pre-deployment evaluation of tools. But principles 1 and 2 are the ones that change what happens when a teacher suspects AI.

Why detectors can’t decide alone: how they work and fail

An AI-text detector is a classifier that estimates how likely a passage is to have been produced by a language model. The common approach measures how statistically predictable the words are. Language models tend to choose likely next words, so their text is often smooth and regular. The detector compares a student’s essay to that pattern and outputs a probability.

That design has a built-in weakness. Predictable, regular writing isn’t unique to machines. Students taught to use standard structures, transition words and formal vocabulary write that way. So do many students writing in a second language.

The independent research is consistent:

• A 2023 study in the International Journal for Educational Integrity by Debora Weber-Wulff and colleagues tested 14 detection tools and found them “neither accurate nor reliable,” all below 80% accuracy. The authors concluded the tools are unsuitable “for use as evidence of academic misconduct” and that paraphrasing or machine translation can easily fool them.
• A 2023 Stanford study published in Patterns by Weixin Liang and colleagues found GPT detectors “consistently misclassify non-native English writing samples as AI-generated.”

Vendors report better accuracy for their current products. Those are the vendors’ own numbers.

There’s also a math problem that any engineer recognizes. Even a small false-positive rate produces real accusations at scale. Take a hypothetical detector that wrongly flags 1 in 100 honest essays. A school that runs 10,000 honest essays a year through it would flag about 100 students who did nothing wrong. A detector score is a signal that something might deserve a look, not a verdict. Principle 2 writes that into policy.

And it reflects a cost researchers have documented. Harvard’s Center for Digital Thriving reported in September 2026 that suspicion has created “two-way suspicion” between teachers and students, and that some students now deliberately do worse work to avoid being flagged. We cover that finding in our article on cognitive withholding.

What the policy gets right

It separates the rule from the evidence. Undisclosed AI use is a real violation. But proving it requires more than software. That’s how fair processes work everywhere else: the rule is clear, and the evidence standard is real.

It names better evidence. “Educator evaluation of student engagement, alternative assessment, and documented evidence of authorship” is a practical list. A teacher who knows how a student writes, an oral follow-up where the student explains their argument, and drafts or version history are far stronger than a percentage.

It gives families a way to respond. The policy calls for “clear criteria for rebutting claims of inappropriate AI use made available to students and parents/guardians.” That’s the part most school policies lack: not just “we can accuse,” but “here’s how you answer.”

It makes disclosure normal, not shameful. By expecting teachers to disclose their own AI use too, it frames disclosure as ordinary practice. That matters because, as Stanford researchers Denise Pope and Victor Lee have noted, most students think some AI help is fine and having AI write a whole paper is not. A disclosure rule lets the honest middle exist out loud.

Where it’s thin

It’s voluntary. Your district may not have adopted it. Ask.

“Clear disclosure” isn’t defined. The policy doesn’t say what a disclosure looks like: a sentence at the end? A citation? A log of prompts? Teachers will need to specify. Our guide to how kids can cite AI in homework offers practical formats.

Rules that vary by assignment are confusing. Principle 5 is right that teachers should decide, but six classes with six AI rules is a lot for a 12-year-old to track. The policy asks teachers to communicate expectations clearly; families can help by asking for them in writing.

Detectors may still be used. The policy limits detectors to one input among several; it doesn’t ban them. A flag can still start a process, so the process needs to be good.

What parents can do with this policy

Find out if your district adopted it

Search your district’s board policies for “artificial intelligence,” or ask the principal directly. If the district adopted a version, ask for the exact text of the disclosure and detector provisions.

Teach disclosure as a habit

Agree on a simple format at home: a line at the end of any assignment where AI was used, such as “I used ChatGPT to brainstorm three topics and to check my grammar. The argument and writing are mine.” Kids who disclose routinely are protected by exactly the rule that catches kids who don’t.

Keep a process trail

Drafts, outlines, notes and version history are “documented evidence of authorship” in the policy’s own words. Encourage writing in tools that track history. If your child is ever questioned, that history is the strongest answer.

Know how to respond to an accusation

If a teacher flags your child’s work, ask calmly what evidence beyond a detector score exists, request the chance for your child to explain the work, bring the drafts, and ask for the district’s rebuttal process. The Turnitin lawsuit coverage shows how high the stakes can get when that process is missing.

What not to do: tell your child that disclosure is risky

Some kids hide AI use because they fear any mention will be held against them. Under this policy, the opposite is true: undisclosed use is the violation. Make disclosing the safe default in your house.

What to Watch For Over the Next 3 Months

• Week 4: Ask each of your child’s teachers, or check syllabi, for the AI rule in that class. Write them down together.
• Month 2 red flags: A grade penalty based only on a detector score, no explanation of the AI rule for an assignment, or your child saying they avoid mentioning AI because “you get in trouble.”
• Month 3 self-check: Does your child disclose AI use without being reminded? Do they keep drafts? Do you know your district’s process for disputing an AI accusation?

Frequently Asked Questions

What is California’s model AI policy for schools?

It’s a 26-principle template from the California Department of Education, required by Senate Bill 1288 (2024), that districts and charter schools can adapt. It covers plagiarism and disclosure, limits on AI detectors, privacy, vendor contracts, AI literacy and grading. The CDE says it is exemplary and compliance is not mandatory.

Does the policy ban AI detectors?

No. It says detectors “should not be used as the sole basis for disciplinary action or grade penalty.” Concerns about AI misuse should rest on educator evaluation, alternative assessment and documented evidence of authorship, with clear ways for students and parents to rebut claims. A detector can prompt a look, not decide the outcome.

Is using AI on homework plagiarism under this policy?

Not automatically. Teachers decide what AI use is allowed on each assignment. What the policy calls plagiarism is submitting AI-generated content “without clear disclosure.” Using AI within the teacher’s rules and saying how you used it is consistent with the policy.

What counts as evidence my child wrote their own work?

The policy points to “documented evidence of authorship” and educator judgment. In practice that means drafts, notes, version history, a teacher’s knowledge of your child’s writing, and your child’s ability to explain the work in conversation or an alternative assessment.

My child isn’t in California. Does this matter to us?

Yes, as a model. Many states and districts are writing AI policies now, and California’s pairing of a clear disclosure rule with limits on detectors is a useful template. You can ask your own district whether its policy treats detector scores as proof or as one signal among several.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. California Department of Education. (2026, last reviewed June 26). “Model Policy: Artificial Intelligence in Education.” https://www.cde.ca.gov/ci/pl/aipolicy.asp
2. Weber-Wulff, D., et al. (2023). “Testing of detection tools for AI-generated text.” International Journal for Educational Integrity, 19, 26. https://link.springer.com/article/10.1007/s40979-023-00146-z
3. Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). “GPT detectors are biased against non-native English writers.” Patterns, 4(7). https://doi.org/10.1016/j.patter.2023.100779
4. Tench, B., Weinstein, E., & James, C., with Starks, A., & Konrath, S. (2026). “An AI Policy Isn’t a Playbook.” Center for Digital Thriving, Harvard Graduate School of Education. https://digitalthriving.gse.harvard.edu/an-ai-policy-isnt-a-playbook/
5. Stanford Graduate School of Education. (2023). “What do AI chatbots really mean for students and cheating?” https://ed.stanford.edu/news/what-do-ai-chatbots-really-mean-students-and-cheating