Migliaccio & Rathod LLP is investigating whether Cursor users’ source code, prompts, editor activity, code snippets, and other programming data may have been used to train or improve artificial-intelligence models without users clearly understanding or affirmatively agreeing to that use.
Cursor is an AI-powered code editor developed by Anysphere. Developers may use it while working with proprietary source code, personal projects, commercial software, client work, or other valuable technical material.
Cursor’s current policies state that when Privacy Mode is enabled, customer data is not used for training by Cursor or its model providers. When Privacy Mode is turned off, however, Cursor states that it may use and store codebase data, prompts, editor actions, code snippets, and other code data and actions to improve its AI features and train its models.
What We Are Investigating
Migliaccio & Rathod LLP is investigating how Cursor’s AI-training practices and privacy settings changed over time, including whether users who supplied code under earlier versions of Cursor’s policies understood that their programming activity could be used to improve AI models.
We are investigating whether:
- Cursor clearly disclosed when user code could be used for AI training;
- Privacy Mode was turned off by default for certain individual users;
- users understood the consequences of leaving Privacy Mode disabled;
- older source code, prompts, or editor activity became eligible for AI training before current disclosures were adopted;
- users were clearly told what categories of programming data could be collected;
- changing privacy settings stopped only future use or also affected information already used for model development;
- Cursor adequately distinguished between processing code to provide an AI feature and using that same code for broader model training; and
- users were provided meaningful control before commercially valuable source code was used for a separate AI-development purpose.
Cursor currently states that users who disable Privacy Mode may allow Cursor to use codebase data, prompts, editor actions, code snippets, and other code-related information to improve AI features and train its models.
Cursor’s Current Training Terms
Cursor has recently adopted particularly clear language regarding AI training.
Its current Terms of Service state that Anysphere will not use a user’s content—or permit a third party to use it—to train AI models unless the user has explicitly agreed to that use.
Cursor’s Privacy Policy similarly states that user inputs and AI-generated suggestions generally are not used for training unless the user has expressly agreed, submitted the material as feedback, or the material is reviewed for certain security purposes.
These newer protections raise questions about how content was handled before explicit-agreement language became part of the governing terms.
Why Source Code Matters
Source code can be economically valuable and commercially sensitive.
A developer may provide code to Cursor because the software needs access to that code to answer a question, generate a suggestion, debug a program, or perform another coding task.
That does not necessarily mean the developer expects the same code to be retained and reused as training material for a broader commercial AI system.
The distinction can be particularly important for:
- independent developers;
- freelancers;
- startup founders;
- open-source developers;
- contractors;
- developers working on proprietary applications; and
- individuals writing code containing original algorithms or commercially valuable functionality.
Potential Claims May Include:
- unfair or deceptive trade practices;
- breach of contract or related contractual claims;
- unjust enrichment;
- inadequate disclosure concerning use of source code for AI development;
- privacy-related claims; and
- claims concerning commercial reuse of valuable user-created code.
Signs You May Be Affected:
- You used Cursor before its current explicit-consent training terms were introduced;
- you used an individual Cursor account rather than an Enterprise account;
- you worked with original or proprietary source code in Cursor;
- you entered substantial prompts, code snippets, or other programming material into the editor;
- Privacy Mode was turned off while you were using Cursor;
- you did not understand that disabling Privacy Mode could allow your code or editor activity to be used for AI training;
- you later enabled Privacy Mode after learning how the data could be used;
- you do not remember affirmatively consenting to use of your code for AI training; or
- you have older screenshots, emails, privacy notices, or Terms of Service showing how Cursor described its data practices at the time.
This investigation is particularly interested in individual Cursor users who used the service before the current explicit-consent language was adopted. Cursor now states that training requires explicit agreement, while its current documentation provides strong protections when Privacy Mode is enabled.
If you used Cursor and believe your source code, prompts, or programming activity may have been used to train AI models without your informed consent, we would like to hear from you.
Please complete the contact form on this page, send us an email at [email protected], or give us a call.