Berkeley's approach to AI comes down to two things: your professor decides, and the Code of Student Conduct covers what happens if you ignore them. In between, the campus gives you a set of licensed tools and some practical advice that's worth following.
Here's what the official guidance says, and which models fit the work Berkeley students do in EECS, data science, Haas and the humanities.
What UC Berkeley actually says
The most useful student-facing page is AI tools for Berkeley students on the campus Life site, published February 24, 2026. We checked it in October 2026. It covers three things.
Your instructor sets the rules. Professors can set strict AI rules for their own courses, so the page tells you to check each syllabus and ask rather than assume. Using AI in a way the syllabus prohibits may be considered academic misconduct.
The Code of Student Conduct applies. Unapproved use falls under section 102.01 of the Code, the same section that covers cheating, plagiarism and fabrication. There's no separate AI rulebook. Unapproved AI help is treated like any other unapproved help.
What you get with a berkeley.edu account. The page lists Google Gemini, NotebookLM and Zoom AI Companion, which needs a short AI Essentials training first. Students with a note-taking accommodation through the Disabled Students Program can get Glean for AI note-taking. The current list lives on the Berkeley AI Hub.
The page also gives sensible advice: use AI as a tutor rather than to finish homework outright, verify facts before you cite them, and keep track of your AI use so you can cite it where it's allowed. Use should also be consistent with Berkeley's Principles of Community and the UC Principles of Responsible AI.
What that means in practice
- There is no campus-wide yes or no. Two classes in the same week can have opposite rules.
- The penalty framework is the usual one. Unapproved AI use is handled like any other academic dishonesty.
- Track your use as you go. If a course allows AI, you'll need to say how you used it. A running note is easier than reconstructing it at midnight.
- Use campus accounts for course material. Gemini and NotebookLM through your berkeley.edu account are the campus-supported route.
Which models suit Berkeley coursework
EECS and CS
Berkeley's CS courses lean on large projects and autograders, and many set their own AI rules, so read them first. Where AI is allowed for learning, use it to understand, not to generate the solution.
Try: "I'm getting a recursion depth error in this Python function. Don't fix it. Ask me questions that lead me to find the base case problem myself."
Claude and ChatGPT are both strong at explaining code. DeepSeek is a strong reasoning model worth adding for algorithm questions. Our DeepSeek vs ChatGPT page compares them.
Data science
Data science is one of Berkeley's signature programs. AI is useful for understanding what a method is doing.
Try: "Explain what a permutation test is testing, in plain language, using an example about two groups of plant heights. Then tell me the most common mistake students make when interpreting the p-value."
Haas and business classes
For case prep, use AI to pressure-test your thinking, not to write your memo.
Try: "Here's my recommendation for this case and my three reasons. Argue the opposite position as strongly as you can, then tell me which of my reasons is weakest."
Grok and Claude give quite different styles of pushback, which is useful for this. Sending the same prompt to both is a quick way to see it.
Humanities and design
For essays, ask for critique of your own draft. For studio or design classes, image models are good for exploring directions quickly. Our Nano Banana vs FLUX for design students post compares two of the best.
How to use AI at Berkeley without breaking the rules
- Check every syllabus, every term. A course you took last year may have changed its policy.
- Ask before you assume. The campus page says exactly this. A short email to the instructor or GSI settles it.
- Use AI as a tutor. Explanations, practice questions and feedback on your own work are the uses Berkeley itself recommends.
- Verify before you cite. Open every source. Models invent references.
- Log your use. Tool, date, what you asked and what you kept.
- Keep sensitive data out of public tools. Student records, unpublished research and anything confidential belong only in tools the campus approves for that data.
Where LazySusan fits
Gemini and NotebookLM through your Berkeley account are a good default for course material. LazySusan is useful when you want models the campus doesn't provide, or want to compare them: personal projects, side work, a startup idea, or checking a tricky answer.
It brings ChatGPT, Claude, Gemini, Grok, Perplexity and DeepSeek into one workspace with one bill, along with image and video models. Multi-Chat sends one prompt to several models at once so you can read the answers side by side. If you're deciding between two models for a class, our ChatGPT vs Claude for students post is a good start.
The Student plan is $99 a year with a .edu email, with 5,000 tokens a month that roll over. Prompt School teaches prompting basics, and you can sign up here.
LazySusan isn't affiliated with or endorsed by UC Berkeley. Your syllabus and the Code of Student Conduct come first.
The short version
At Berkeley, each professor decides, and the Code of Student Conduct handles the rest. Use the campus tools for course material, treat AI as a tutor, verify everything and keep a record of how you used it.