Eigma
EN 中 DE
What Is Inside a Chatbot Eigma öffnen
Inhalt Privacy and what you send

Living with it

Privacy and what you send

What happens to what you type is decided by that product's own terms: not free against paid, but consumer app against API and business plan.

On the ChatGPT and Claude consumer plans, training on your chats is a setting someone has to set, and the two vendors default it differently. Anthropic's August 2025 consumer terms update makes it an active choice on Claude Free, Pro and Max: allowing it extends retention of new and resumed chats to five years, declining leaves the existing thirty days in place, and existing users had until 8 October 2025 to choose. Claude for Work, Government, Education and API access are excluded from that change.1 Deleting a chat is also not erasure: it leaves the visible history at once and can persist on back-end systems for up to thirty days.2 OpenAI documents its consumer ChatGPT plans as using content to improve models unless the user opts out, while ChatGPT Business, Enterprise, Edu and the API are not used for that by default.3

Product Used to train by default Retention
ChatGPT Free and Plus Yes unless you opt out3 Not stated on the pages cited here
ChatGPT Business, Enterprise, Edu No3 Not stated on the pages cited here
OpenAI API No without an explicit opt-in, since 1 March 20234 30 days of abuse monitoring; zero retention by approved application4
Claude Free, Pro, Max Only if you allow it1 30 days, or five years if you allow training12
Claude API Not retained by default, and never trained on without express permission5 Designated models require 30 days; uploaded files until deleted or expired5
Gemini in Workspace apps No6 Not stated on the page cited here
Gemini consumer apps No primary policy page obtained No primary policy page obtained

The developer terms are written differently. OpenAI has documented since 1 March 2023 that API data is not used to train its models without opt-in, and that reduced monitoring or zero retention is available only by approved application.4 Anthropic's API documentation states that conversation content is not retained by default and is not used for training without express permission.5 Individual features carry their own terms regardless: uploaded files stay until deleted or expired, code-execution container data up to thirty days.5

Two cautions. These are dated policies, not properties of the technology, and the Anthropic consumer data-retention page cited here carries a July 2026 date; not every page cited carries that date. And no vendor publishes its consumer and business terms side by side, so the table was assembled from several pages, and the consumer Gemini row stays empty rather than guessed. When in doubt which column you are in, assume the consumer one.

Quellen

  1. [1]Anthropic, updates to our consumer terms
  2. [2]How long do you store my data?
  3. [3]OpenAI, how your data is used to improve model performance
  4. [4]OpenAI API, your data
  5. [5]Claude API and data retention
  6. [6]Gemini in Google Workspace, your data
Quizze

  1. Someone is deciding whether a piece of sensitive text is safe to paste in. Which distinction in the published terms should drive that decision?

    • Consumer app terms against API and business terms
    • Whether the free plan or the paid plan is in use
    • Whether the app is opened on a phone or a laptop

    Business and developer terms document no training by default. Consumer plans document training as a setting, opt-out for one vendor and an active choice for another.

  2. Paying for a consumer chatbot subscription automatically puts that account under the same no-default-training terms documented for business and API plans.

    • True
    • False

    Claude Pro and Max are paid consumer plans, yet training is still a choice the user must set there, unlike Claude for Work, Education or API access, which are excluded from that choice.

Kommentare

Noch keine Kommentare. Fang das Gespräch an.