Policies
The ethics pledge, the AI policy — what tools you can use and how to disclose them — and the University-required statements. The remaining course rules (grading, late work, collaboration) are in the syllabus.
Ethics pledge
You must sign the ethics pledge in A0 before you receive the course materials and VM access. It has three rules: do not alter another person’s programs or data; do not access systems you are not authorized to access; do not use resources you are not authorized to use. Everything you learn here is constrained by those three sentences. Techniques from this course applied outside the course environment, without written authorization, are a University matter and possibly a criminal one.
AI policy
The short version: use any tool you want on the code — just disclose it. Your understanding is checked in a short oral, in a room, with no tools. Disclosure is never penalized; non-disclosure is the violation.
Every graded artifact carries one of two labels. The label is printed on the handout. If you’re unsure which applies, ask on Canvas Discussions before you start.
AI-OK
Applies to: all assignments (A0–A6) and in-class labs.
Use whatever you want — Claude, ChatGPT, Copilot, Cursor, a decompiler plugin, a friend’s blog post, Stack Overflow. There is no restriction on tools.
One requirement: submit an AI-USE.txt with your solution (template below). Name the tools and say
what they did. That’s it — the point is attribution, not surveillance.
What counts as adequate disclosure:
- Name each AI system you used and describe in a sentence what you used it for. That is the whole bar.
- You do not need to submit prompts or whole conversations — unless the assignment asks for a transcript, which A2, A3 and A6 do (see below).
- You do not need to log editor autocomplete, grammar or spell check, search results, or a lookup you would have done anyway.
- Generating the entire solution with AI is allowed. Say that you did. The oral is where your understanding gets checked, and that is the whole design.
“I asked Claude to explain what
%ndoes and it wrote my first draft of the payload” — costs you nothing. Failing to say so is an honor offense.
A few assignments also ask for the transcript. For the “beat the model” parts of A2 and A3 and the A6 tournament, your AI session transcript is part of the deliverable and is graded — we want to see what you asked, what came back, where it was wrong, and what you did about it. A transcript showing you caught the model’s mistake is worth more than one showing it got everything right.
NO-AI
Applies to: all five oral-evaluation rounds, the in-class quizzes, and the final exam.
In the room, no network and no devices beyond whatever we provide for the session — the quizzes and the final exam are written on paper. Nothing to disclose because nothing is permitted.
Why it’s built this way
I can't detect AI use. Nobody can — the detectors don't work, and I'm not going to make you sign something we both know is unenforceable.
So here's the deal instead. On the code, use anything you want. Use the best tools available, because that's what you'll do at work, and the code is where you learn.
But three quarters of your grade — the five oral evaluations, the quizzes, and the final exam — is earned in this room with no network. The orals are the heart of it: five five-minute conversations about your own code, one per assignment. I'll ask you to change something and tell me what happens.
If you did the assignment, that's a conversation you'll enjoy. If something else did it, you'll find out in Week 4 instead of in a job interview three years from now. That's not a threat, it's the whole design. I'd rather you find out here.
And one more thing: a whole module of this course is about attacking and defending AI systems. Banning AI in a course that teaches you to break it would be a little absurd.
Enforcement, honestly
| Situation | Response |
|---|---|
| Heavy AI use, fully disclosed | Fine. No penalty. This is the policy working. |
| AI use, not disclosed, discovered | Honor referral. The violation is the non-disclosure. |
| Code near-perfect, oral near-zero | Not an honor case. It's a grade — the oral half is worth as much as the code half, and it handles this on its own. |
| Viva reveals you can't explain your own submitted code | Score the viva accordingly (0–2). Still not an accusation. |
| Any device or network in a NO-AI session | Standard exam-integrity procedure. |
What this policy deliberately does not do
- No AI-detection tools. They don’t work, they produce false accusations, and they fall hardest on non-native English speakers.
- No commit-history or “show your process” forensics as evidence. Cheaply fabricated.
- No honor-pledge theater. One ethics pledge in A0, meaning what it says, is worth more than a signature block on every submission.
AI-USE.txt template
Ship this in every assignment submission.
# AI Use Disclosure — A<N>, <your name / computing ID>
## Tools used
<!-- e.g. Claude via claude.ai; GitHub Copilot in VS Code; none -->
## What I used them for
<!-- one or two sentences per tool. Be specific:
"explained the %n format specifier",
"wrote the first version of gen_hacked_input.sh which I then fixed" -->
## What I did myself
<!-- the part you'd be able to reproduce on a whiteboard -->
## Where the tool was wrong
<!-- optional, but genuinely interesting; helps on borderline grades -->
## Other sources
<!-- URLs, classmates you discussed concepts with, etc. -->
University policies & statements
Academic integrity
Students are expected to be familiar with the UVA honor code, including the section on academic fraud. Unless otherwise noted, exams and individual assignments are pledged: you have neither given nor received unauthorized help. Consulting solutions or exams from previous years is a violation. If you have questions about what is allowable, ask.
Students with disabilities
It is my goal to make this course as accessible as possible. If you anticipate any barrier related to the format or requirements of this course, meet with me, and consider working with the Student Disability Access Center (SDAC). If you are already approved for accommodations, send me your letter and we’ll build an implementation plan together. SDAC accommodations apply to oral evaluations, quizzes, and the final exam.
Religious accommodations
It is University policy to reasonably accommodate students so that they do not experience an adverse academic consequence when sincerely held religious beliefs or observances conflict with academic requirements. Submit any request to me by email as far in advance as possible. Questions may go to the Office for Equal Opportunity and Civil Rights.
Harassment, discrimination, and interpersonal violence
UVA is dedicated to a safe and equitable learning environment. Information and resources are on the UVA Sexual Violence site. As a faculty member I am a responsible employee and required to report certain disclosures to the Title IX Coordinator, whose role is to ensure you get resources and support. For a confidential resource, contact CAPS or the Women’s Center.
Ethics statement
This course covers technologies whose abuse could infringe on the rights of others. As computer scientists we rely on the ethical use of these technologies. Circumventing security or privacy mechanisms, or disseminating or exploiting vulnerabilities outside the letter and spirit of these guidelines, will be reported to the proper authorities and may result in dismissal from the class and further academic and legal sanctions.
Career development
I want to support your career development — a research conference, an interview, an externship. Notify me by email as far in advance as possible to arrange accommodations.