Does OpenAI train on your data and who answers for what a chatbot says?

OpenAI says API data has not been used for training since 1 March 2023, but keeps abuse logs up to 30 days; free and paid individual ChatGPT plans may be trained on unless you opt out. A BC tribunal ordered Air Canada to pay $812.02 for its chatbot's wrong advice. Read 2026-11-06.

By Mackenzie Bowes · mackenzie@mackenziebowes.com · @_mackenziebowes on X

Contents

Two questions come up before a business puts an AI tool in front of customer or client information. What does the vendor do with what is typed in, and who answers when the tool says something wrong? This page answers the first from OpenAI's own pages and the second from the one tribunal decision that ruled on a company's chatbot. It does not cover other vendors, and it is not legal advice.

Does OpenAI train on API data?

For the API, no, by default. OpenAI's developer guide says: "Your data is your data. As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)."[1] Its business privacy page extends the default to ChatGPT Business, ChatGPT Enterprise and the API Platform: "By default, we do not use your business data for training our models", unless a customer opted in, for example through feedback mechanisms.[2]

Not training is not the same as not keeping. The guide says "abuse monitoring logs are generated for all API feature usage and retained for up to 30 days, unless longer retention is required by law", and that those logs "may contain certain customer content, such as prompts and responses".[1] Customers can ask to have content left out of those logs through Zero Data Retention or Modified Abuse Monitoring, but "these controls are subject to prior approval by OpenAI and acceptance of additional requirements."[1] Zero Data Retention also changes endpoint behavior, and some endpoints "may still store application state".[1]

The individual plans differ

OpenAI's help article says the rule changes on the plans for individuals: "When you use our services for individuals (Free, Plus, Pro), we may use your content to train our models." To opt out, a user turns off "Improve the model for everyone" under Settings > Data controls.[3] Giving feedback with a thumbs up or down overrides an opt-out for that conversation: "the entire conversation associated with that feedback may be used to train our models."[3] Temporary chats "are not used to improve OpenAI models while they remain temporary".[3] So an employee pasting client records into a personal ChatGPT account is under different terms from the same employee on the business plan or the API. The pages read do not say which plan a given employee is on; that is the business's own record to check.

Who answers for a chatbot's wrong answer?

In Moffatt v. Air Canada, 2024 BCCRT 149 (issued February 14, 2024), a British Columbia Civil Resolution Tribunal member decided a small claim over a chatbot on the airline's website. Jake Moffatt asked about bereavement fares. The chatbot said a reduced fare could be claimed after travel, within 90 days; Air Canada's staff later said it did not allow retroactive applications.[4] The claim was $880 for the fare difference.[4]

Air Canada argued it could not be held liable for information from its chatbot. The tribunal member wrote: "In effect, Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions. This is a remarkable submission." And: "It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot."[4] The decision found negligent misrepresentation: "I find Air Canada did not take reasonable care to ensure its chatbot was accurate."[4] Air Canada had pointed to a correct webpage elsewhere on its site; the decision says it "does not explain why customers should have to double-check information found in one part of its website on another part of its website."[4] Air Canada also relied on terms in its tariff but did not provide the tariff, and the decision notes that "it is not enough in a legal process to assert that a contract says something without actually providing the contract."[4]

The order: Air Canada to pay a total of $812.02 within 14 days, made of $650.88 in damages, $36.14 in pre-judgment interest and $125 in tribunal fees.[4] The decision is from a small-claims tribunal and says nothing about whether other courts must follow it. For the failures that lead to cases like this, see AI agent failure modes; for what filters can and cannot stop, see AI agent security and guardrails. For the Canadian consent rules that apply to commercial email an automated sequence sends, see Automating follow-up emails with AI: methods, rules.

Not yet proven

If your business uses an AI tool with client information and you have written down the terms you rely on, I would like to compare notes: mackenzie@mackenziebowes.com.

Sources

  1. OpenAI, Data controls in the OpenAI platform. Read in Rotunda 2026-11-06. developers.openai.com/api/docs/guides/your-data
  2. OpenAI, Enterprise privacy at OpenAI. Read in Rotunda 2026-11-06. openai.com/enterprise-privacy/
  3. OpenAI Help Center, How your data is used to improve model performance. Read in Rotunda 2026-11-06. help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance
  4. Civil Resolution Tribunal of British Columbia, Moffatt v. Air Canada, 2024 BCCRT 149 (CanLII), File SC-2023-005609, a copy of the decision as a PDF hosted by a third party. Read in Rotunda 2026-11-06 and converted with IndraOCR. s3.amazonaws.com/IGG/AI+Part+1+-+Materials/Moffatt+v.+Air+Canada.pdf

Terms on this page

  • AI agent: A system where a language model decides for itself how to do a task and which tools to use.