Oral Defenses at Home: UChicago's AI-Resilient Method

On July 9, 2026, the University of Chicago Law School published its AI strategy, and the notable thing isn’t what it banned. It’s what it required. Students must sit down with their supervising professor and defend their research paper out loud, answering questions that probe the paper’s reasoning. No detection software. No surveillance. The school calls the approach “AI-resilient” rather than AI-resistant, and the lead drafter explained the word choice as an attempt to keep the learning intact even when AI is available. An oral defense for kids’ homework is the two-minute version of that, and it works at every grade level.

Key Takeaways

• UChicago Law requires an in-person oral discussion of every Substantial Research Paper with the supervising professor, after a complete draft, where students answer questions probing “the paper’s reasoning and argument implications.”
• The school defines AI-resilient as modes of evaluation that “reward students’ effortful and sustained engagement with the material, and discourage the offloading of work to AI tools.”
• It also bans devices: “Across all 1L sections, we will prohibit the use of electronic devices such as laptops, tablets, and phones in the classroom.”
• Detectors are the alternative and they don’t work. Weber-Wulff et al. (2023) tested 14 tools and found them “neither accurate nor reliable,” biased toward calling AI text human.
• Oral questioning doubles as retrieval practice. Roediger and Karpicke (2006) found testing produced “substantially greater retention than studying.”

What UChicago actually did, and why it’s the smarter move

The strategy statement, titled “Rethinking Legal Education in the AI Era,” came out of a year-long consultation with alumni, law firm leaders, legal technology executives, young associates, faculty, and students. Two mechanisms came out of it.

First, a device ban in the core first-year curriculum: “Across all 1L sections, we will prohibit the use of electronic devices such as laptops, tablets, and phones in the classroom,” with narrow exceptions for designated scribes, professor-authorized activities, and disability accommodations. Nine core courses, from Civil Procedure to Transactional Lawyering.

Second, and more portable to a kitchen table, the oral defense. Every student must have an in-person oral discussion of their Substantial Research Paper with the supervising professor, after submitting a complete draft or final version, either one-on-one or as a workshop-style class presentation. During that session, the student answers questions probing the paper’s reasoning and the implications of its argument.

The school’s own definition of “AI-resilient” is worth quoting because it’s a design principle, not a policy: modes of classroom interaction and performance evaluation that “reward students’ effortful and sustained engagement with the material, and discourage the offloading of work to AI tools.”

Notice what’s absent. No AI detector. No plagiarism scanner. No surveillance software. The school’s insight was that you don’t need to prove a student used AI if you can simply establish whether they understand their own work. And they’re not alone in the direction: Columbia Law’s August 7, 2026 policy permits AI as a learning aid, “such as summarizing notes or generating practice hypotheticals,” while barring it from submitted work, and UC Berkeley Law on May 22, 2026 prohibited generative AI to “conceptualize, outline, draft, revise, translate, or edit any graded work.”

Why detectors are the wrong tool, with receipts

The alternative approach, running work through an AI detector, fails on two counts: accuracy and fairness.

Weber-Wulff, Anohina-Naumeca, Bjelobaba, Foltýnek, Guerrero-Dib, Popoola, Šigut, and Waddington (2023, International Journal for Educational Integrity) tested “12 publicly available tools and two commercial systems (Turnitin and PlagiarismCheck)” and concluded that “the available detection tools are neither accurate nor reliable and have a main bias towards classifying the output as human-written rather than detecting AI-generated text.” They also found that “content obfuscation techniques significantly worsen the performance of tools.”

Liang, Yuksekgonul, Mao, Wu, and Zou (2023, Patterns) found that “GPT detectors frequently misclassify non-native English writing as AI generated,” and argued that “addressing the biases in these detectors is crucial to prevent the marginalization of non-native English speakers.”

And the real-world costs are documented. In Wake County, North Carolina, a Green Hope High School freshman said she was falsely accused after a teacher ran her English assignment through three detection tools returning 62%, 75%, and 87% likelihood of AI use; another teacher reviewed her document version history and confirmed she hadn’t used AI. The district’s subsequent draft policy stated that it “does not support the use of AI detection programs due to their technical unreliability, inaccuracy, and potential for bias against specific student populations, including those for whom English is a second language.” At Purdue, an instructor emailed more than 200 students in a CS 240 course alleging AI use; the allegations were dropped within days.

An oral defense has none of these failure modes. It doesn’t produce false positives, because it isn’t a verdict about tool use. It’s a measurement of understanding.

The question bank

Group these by what they test, not by subject. Pick two or three per session. Never more.

What it tests	Grades 3–5	Grades 6–8	Grades 9–12
Understanding the core idea	”Tell me what this is about in one sentence."	"If a friend missed this unit, what’s the one thing they’d need to know?"	"What’s the strongest version of the opposing view?”
Process	”What did you do first?"	"Where did you get stuck, and what did you try?"	"What did you cut, and why?”
Reasoning	”Why does that step work?"	"What would happen if you changed this number?"	"Which part of your argument is weakest?”
Sourcing	”Where did you learn that?"	"Which source did you trust most and why?"	"Did you verify this claim independently? How?”
Transfer	”Can you give me a different example?"	"Where else would this rule apply?"	"How would this change in a different context?”
Ownership of language	”What does this word mean?"	"You used the word [X]. Say it another way."	"Explain this sentence without using any of its words.”
Error-finding	”Is there anything here you’re unsure about?"	"Find me one thing you’d fix with more time."	"What’s the assumption you didn’t state?”
AI disclosure	”Did anything help you with this? What part?"	"What did you use AI for, and what did you do yourself?"	"Which sentences are yours, and which are edited?”

The last row deserves a note. Ask it openly and without an accusatory tone, and kids mostly answer honestly. Ask it as a trap and they won’t. The purpose isn’t to catch them; it’s to make disclosure a normal, low-stakes part of finishing work, which is exactly what Katy ISD’s framework requires of its students, who must “verify AI-generated information and disclose or cite AI use when appropriate.”

The “ownership of language” row is the most diagnostic. A student who used AI heavily can usually summarize the content but cannot rephrase a specific sentence, because the sentence was never theirs. You don’t have to say anything about it. You’ll just know, and the right follow-up is “let’s rewrite that one together” rather than an accusation.

Running it at home without making it a tribunal

Two minutes, after the work, every time

Not a test. A close. “Before you put that away, walk me through this part.” Two minutes, two questions, same rhythm each time. The predictability is what makes it stop feeling like an interrogation. Within a couple of weeks it becomes as unremarkable as putting the dishes in the sink.

Ask questions you don’t know the answer to

The best defense questions are genuine. “Which part was hardest?” is a real question with an unknown answer. “Did you actually do this yourself?” is not a question, it’s an accusation with a question mark. Genuine curiosity gets genuine answers, and it also means you learn something about how their brain works.

Let them defend a right answer

Run the defense when the work is correct, not only when you’re suspicious. That’s the whole trick. If the defense only appears when you doubt them, it becomes a signal of distrust. If it happens every time, it’s just how work ends in your house.

Use it instead of, not alongside, detection tools

Don’t do both. Running an oral defense and also checking their work against a detector undermines the entire premise, which is that understanding is the thing you care about. Given what Weber-Wulff’s team found, the detector is also the less accurate of the two methods.

Make the retrieval count

This isn’t just an integrity mechanism, it’s the best-supported study activity there is. Roediger and Karpicke (2006, Psychological Science) found that “repeated testing produced substantially greater retention than studying,” and that students felt more confident after restudying even though testing worked better. Two minutes of oral questioning does double duty, which is why it’s worth doing even on work you have no doubts about.

What not to do

Don’t run it in front of siblings. The defense only works if getting something wrong is safe, and an audience makes it unsafe. Also don’t do it at bedtime; a tired kid performs worse and you’ll misread the result.

What to Watch For Over the Next 3 Months

• Week 4: Has the defense become routine, or does it still trigger defensiveness? Routine is the goal, and it usually arrives around week three.
• Month 2 red flags: They can summarize but can’t rephrase specific sentences. That’s the clearest signature of borrowed language, and it needs a rewrite session, not a confrontation.
• Month 3 self-check: Ask them to defend something they wrote a month ago. Retention is the real measure, and it’s the thing AI shortcuts quietly cost.

Frequently Asked Questions

What did UChicago Law actually require?

An in-person oral discussion of every Substantial Research Paper with the supervising professor, held after a complete draft or final version, one-on-one or as a workshop presentation, where the student answers questions probing the paper’s reasoning and the implications of its argument. The school also banned devices from all core first-year classes.

Why not just use an AI detector?

Because they don’t work reliably. Weber-Wulff et al. (2023) tested 14 tools and found them “neither accurate nor reliable,” with a bias toward classifying AI text as human. Liang et al. (2023) found detectors “frequently misclassify non-native English writing as AI generated.”

How long should a home oral defense take?

Two minutes, with two or three questions. Longer turns it into an interrogation. The goal is a routine close to homework, not an examination.

What’s the single most revealing question?

“Explain this sentence without using any of its words.” A student who wrote it can rephrase it; a student who pasted it usually can’t. It’s diagnostic without being accusatory, and the right response to a failure is to rewrite that sentence together.

Won’t my kid feel like I don’t trust them?

Only if you run it selectively. Do it every time, including when the work is clearly their own and clearly correct, and it becomes a normal part of finishing homework rather than a signal of suspicion.

Is this just for older kids?

No. The question bank scales down to third grade: “tell me what this is about in one sentence,” “what did you do first,” “where did you learn that.” The mechanism, demonstrating understanding out loud, is age-independent. Only the questions change.

---

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. University of Chicago Law School. (July 9, 2026). “Rethinking Legal Education in the AI Era.” https://www.law.uchicago.edu/news/ai-strategy-statement
2. Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S., et al. (2023). “Testing of detection tools for AI-generated text.” International Journal for Educational Integrity, 19(26). https://doi.org/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. Roediger, H. L., III, & Karpicke, J. D. (2006). “Test-enhanced learning: Taking memory tests improves long-term retention.” Psychological Science, 17(3). https://doi.org/10.1111/j.1467-9280.2006.01693.x
5. WRAL. (May 5, 2026). “Wake County student says clear AI policies needed after being accused of cheating.” https://www.wral.com/news/education/wake-county-student-says-ai-policies-needed-after-cheating-accusation-may-2026/
6. WRAL. (June 17, 2026). “No AI detectors, more citations: what’s in Wake schools’ AI policy draft.” https://www.wral.com/news/education/whats-in-wake-schools-new-ai-policy-draft-june-2026/
7. Purdue Exponent. (April 21, 2026). “AI panic causes campus uproar.” https://www.purdueexponent.org/campus/general_news/ai-panic-causes-campus-uproar/article_fdb604d1-206a-4883-bbe1-fe7631a2e083.html
8. Covering Katy News. (Aug 11, 2026). “Katy ISD’s new AI rules: What parents and students need to know.” https://coveringkaty.com/education/katy-isd-s-new-ai-rules-what-parents-and-students-need-to-kn/
9. Common Sense Media. (Aug 18, 2026). “Teens in the AI Era: Schoolwork and Skills That Matter.” https://www.commonsensemedia.org/research/teens-in-the-ai-era-schoolwork-and-skills-that-matter

Related reading: AI cheating detectors are failing students, what AI-proof homework looks like now, and a weekly ritual for verifying AI math answers.