Exam prep · Updated 10 September 2026

How to Study for the CFA Exam With AI (Safely)

From February 2027 the CFA Level I curriculum contains a module on large language models. The rules about what you may say about the exam did not loosen to match.

Yes, you can use AI to prepare for the CFA exam. Nothing in CFA Institute's rules restricts your study tools, and its guidance on Standard VII(A) explicitly says the standard "does not prohibit candidates from discussing nonconfidential information or material with others or in study groups in preparation for an exam." Two things make the CFA different from every other exam we cover, though. The curriculum now teaches AI — a new Level I module introduces financial data science, AI and large language models — so you will be examined on the vocabulary, not just using it. And the confidentiality rule is drawn further out than most candidates realise: it covers not only exam questions but "broad topical areas and formulas tested or not tested on the exam."

What changed for February 2027

CFA Institute announced the 2027 curriculum update on 26 May 2026. The new curriculum "applies to candidates sitting exams beginning February 2027" and the changes "represent about 25 percent of the CFA Level I exam," concentrated in two topic areas.

Prep publishers reading the official curriculum-change document report further Level I detail the press release does not spell out: the GIPS reading dropped from Ethics, the seven Standards of Professional Conduct presented as separate modules, and multi-factor models moving down from Level II. Reported rather than official — confirm against your own curriculum access.

The practical consequence is blunt: if you sit Level I from February 2027, roughly a quarter of what you are studying is new, and every chatbot, forum thread and prep summary describing "the CFA quant syllabus" was written about the version before it. Prep publishers list the February 2027 Level I window as 22–28 February with a final registration deadline of 5 November 2026 — confirm both on cfainstitute.org.

The rule that catches CFA candidates out

This is the part that makes the CFA different, and almost no AI study guide mentions it. Standard VII(A) reads: "Members and candidates must not engage in any conduct that compromises the reputation or integrity of CFA Institute or the CFA® designation or the integrity, validity, or security of CFA Institute programs."

The guidance then gives examples of information candidates may not disclose. Two of them:

• "specific details of questions appearing on an exam"
• "broad topical areas and formulas tested or not tested on the exam"

Read the second one twice. It is not just the questions. Saying "there was barely any derivatives this window" is covered. So is "they didn't test Bayes." CFA Institute's separate exam misconduct policy lists both directions of the same act: "Disclosing, verbally or in writing, the content of the exam, in whole or in part" and "Requesting others to disclose to you the content of the exam, in whole or in part." It states that an investigation "may lead to the voiding of your exam results and suspension or termination of your candidacy."

Now map that onto the two most natural things a candidate does with a chatbot. Walking out of the test centre and typing "here is what I remember being asked, help me work out if I got them right" is disclosing exam content in writing. And feeding a model a scraped forum thread of other candidates' recollections so it can build you a practice set is soliciting content that only exists because someone else broke the same rule. Neither feels like cheating in the moment. Both are named in the policy.

There is a small irony here worth using: Ethics is the largest single topic at Level I, and Standard VII(A) is itself examinable. The rule governing how you may study is on the syllabus you are studying.

Level I is a 90-second race, so practise like one

CFA Institute's Level I page states the exam is "180 multiple choice questions" across "two 135-minute sessions" of "90 multiple-choice questions" each, and that this works out to "approximately 90 seconds for each multiple-choice question." All questions are equally weighted and there is no penalty for a wrong answer. You must also "complete one PSM at each level in order to receive your exam result," and successful candidates "report spending over 300 hours on average preparing for each level."

Level III also splits into three specialized pathways — Portfolio Management, Private Wealth and Private Markets — chosen at registration and fixed once you register, sitting on top of a common core. The charter granted is the same whichever you pick.

The 90-second figure is the constraint that should shape how you use AI, and it is the one most study workflows get wrong. A chatbot's default behaviour is a patient four-paragraph explanation. That is excellent for understanding and terrible as a rehearsal, because on exam day nobody explains anything to you and you have a minute and a half. Split the two deliberately: untimed sessions where you make the model explain, and timed sessions where you forbid it to.

The Practical Skills Module problem

The PSMs are the one place where AI can quietly do the whole thing for you. They are hands-on modules — Python, equity analysis, macro insights — and you do not receive your exam result without completing one at your level. A model will write you working Python for a portfolio optimisation in about four seconds.

Do that and you have bought the completion and skipped the skill. If your Python is shaky, our AI workflow for Python and data science covers the honest version: you write the code, the model explains the error.

Four prompts that fit this exam

1. The 90-second drill. The single highest-value prompt here, because pace is what fails candidates who know the material. Set an actual timer.

You are drilling me on CFA Level I [topic] using my own curriculum notes,
which I will paste below.

Rules: one question at a time, three answer choices (A, B, C), no
explanation until I answer. If I take longer than 90 seconds I will type
SLOW - log it and move on.

After 10 questions, give me only: my score, the ones I got wrong, and
which single concept explains most of my errors. Do not teach me anything
until I ask.

Notes: [paste]

2. The ethics scenario generator. Ethics is the biggest topic at Level I and the questions are situational, not definitional. Note the constraint on sourcing: you paste the Standards, the model invents the scenario.

Here is the text of a CFA Standard of Professional Conduct: [paste from
your curriculum].

Invent three short workplace scenarios where a reasonable person might
think this Standard applies. Exactly one should be a genuine violation.
Do not tell me which.

I will name the violation and the specific clause. Then tell me whether I
was right and, for the two non-violations, what would have had to change
to make them violations.

3. The curriculum-anchored explainer. This one exists to stop the model filling gaps with the pre-2027 syllabus.

Explain [concept] to me using only the text I paste below, which is from
my 2027 CFA curriculum.

If the text does not contain something you would normally include, say
"not in your text" rather than adding it from memory. If the text
contradicts what you would have said, follow the text and flag the
difference.

Then ask me two questions to check I actually followed it.

Text: [paste]

4. The Python tutor that refuses to write code. For the PSM.

I am working through a CFA Practical Skills Module in Python. Here is my
code and the error I am getting: [paste].

Do not give me corrected code. Tell me what the error message means in
plain English, point at the line where the problem starts, and name the
concept I have misunderstood. Then let me try again.

All four share one design rule: the model asks, times or critiques, and you produce. That is the same principle behind building Socratic AI tutors for exam prep, and it is what separates studying from watching.

What AI cannot do for you here

• Tell you what is on the exam. Any answer worth having would be a disclosure violation by its source. Any answer it actually gives you is a guess dressed as a fact.
• Give you the current curriculum. About a quarter of Level I moved for 2027. A model trained before that will describe the old Quantitative Methods and Equity Investments confidently and in detail. Paste your own materials in — the habit in our AI hallucination checklist for students.
• Predict whether you will pass. The Minimum Passing Score is set by CFA Institute after the fact and is not published as a percentage. Nothing a chatbot says about your odds is grounded in anything.
• Replace the BA II Plus. You sit the exam with an approved calculator — the Texas Instruments BA II Plus or the HP 12C — and nothing else. Keystroke fluency is a physical skill and only reps build it.

Where your real practice should come from

• CFA Institute's own mock exams and practice questions, which come with your registration. They are the only items calibrated to the real thing; AI-generated questions are calibrated to nothing.
• AI for reps between mocks, using the prompts above. Skill building, never answer supply.
• An error log. One line per miss: topic, what went wrong, the fix. After three weeks it is your revision plan, and it beats any study schedule a chatbot writes you.
• The underlying subject, when the curriculum assumes you already have it. Level I moves fast over material you may have last seen in a lecture hall — our workflows on DCF valuation, CAPM and portfolio theory and balance sheet analysis cover the three that trip people up most.
• The neighbouring credential, if you are weighing both. The accounting route has a comparable set of confidentiality rules that work slightly differently — see how to study for the CPA Exam with AI.

FAQ

Can you use AI to study for the CFA exam?

Yes. No CFA Institute rule restricts what you use while preparing, and the Standard VII(A) guidance states that it "does not prohibit candidates from discussing nonconfidential information or material with others or in study groups in preparation for an exam." The restrictions bite on exam content, not on study tools. Use a model to quiz you, to time you and to argue with your reasoning — and take the actual curriculum from CFA Institute rather than from the model.

What changed in the 2027 CFA curriculum?

CFA Institute announced the update on 26 May 2026. It applies to candidates sitting exams beginning February 2027 and represents "about 25 percent of the CFA Level I exam," concentrated in Quantitative Methods and Equity Investments. The headline addition is a new learning module introducing "financial data science, AI, and large language models" — the first time LLMs appear in the curriculum. Quantitative Methods also expands estimation, simulation and portfolio optimization, and Equity Investments shifts toward modelling, valuation and communicating investment insights.

Is it against the rules to ask a chatbot about questions I saw on the CFA exam?

Yes. CFA Institute lists as misconduct "Disclosing, verbally or in writing, the content of the exam, in whole or in part." Typing what you remember into a chat window is disclosing it in writing. Standard VII(A) guidance goes further and names "broad topical areas and formulas tested or not tested on the exam" as information candidates may not disclose, so even a vague summary is covered. CFA Institute states that an investigation "may lead to the voiding of your exam results and suspension or termination of your candidacy."

Can AI do my Practical Skills Module for me?

Technically it could, and that is the problem. CFA Institute states that you must "complete one PSM at each level in order to receive your exam result," and the modules are hands-on: Python, equity analysis, macro insights. A chatbot can produce working Python in seconds, which gets you the completion and none of the skill the module exists to build. Use AI to explain an error message or to review code you wrote, and write the code yourself.

Can I ask a chatbot which CFA topics to skip?

No, for two separate reasons. First, any answer reflecting what actually appeared on a recent exam could only come from candidates who broke the disclosure rule, and CFA Institute lists "Requesting others to disclose to you the content of the exam, in whole or in part" as misconduct in its own right. Second, the answer would be wrong anyway: about a quarter of Level I changed for February 2027, so a model describing the old content sounds confident and points you at material that has moved.

Bottom line

The CFA is the first major finance credential to put large language models inside its own curriculum, which makes the "is AI allowed?" question easy to answer badly. Preparing with a chatbot is ordinary studying and always was. Saying anything about the exam afterwards is not — and the line sits further out than candidates expect, covering broad topical areas and formulas, not just questions. So: paste your own curriculum in, make the model quiz you at 90 seconds a question instead of lecturing you, write the PSM code yourself, and when you walk out of the test centre, close the app.