Digital Economy

Claude: Anthropic changes rules on user data

The artificial intelligence company revolutionises its privacy policy, extending data retention to five years. 28 September deadline to decide

5 September 2025

2' min read

Anthropic is putting all users of the chatbot Claude in front of a crucial choice: by 28 September they will have to decide whether to allow the company to use their conversations to train artificial intelligence models. This is a radical change that marks a turning point in the strategy of the company founded by ex-OpenAI and comes after ChatGPT has been offering this opt-out possibility since May 2023.

The policy revolution

The change of course is clear. Until yesterday, Anthropic did not use data from consumer conversations for training its models, except on a voluntary basis through user feedback. Now instead, the company wants to use chats and programming sessions, extending data retention to five years for those who do not object. Previously, inputs and conversations were automatically deleted from the servers within 30 days, unless there were legal obligations or violations that could extend retention up to two years.

AI's hunger for data

Anthropic frames the change as a matter of user choice: those who do not object 'will help improve model security' and contribute 'to future Claude models in skills such as programming, analysis and reasoning'.

In essence: help us help you. But the intentions, though painted in an altruistic and noble tone, are of a practical nature. Like any other artificial intelligence company, Anthropic desperately needs data - more than it cares about brand reputation. Training requires huge amounts of high-quality conversations, and millions of interactions with Claude represent the ideal content to strengthen the competitive position against OpenAI and Google.

The problem of informed consent

The implementation of Anthropic's new policy follows a familiar pattern: new users will choose their preference during registration, but existing users are faced with a pop-up with 'Updates to Consumer Terms and Policies' and an 'Accept' button automatically set to 'On'. The concern is that users may quickly accept without noticing that they are agreeing to data sharing. For many experts, the confusion created by these changes is tangible. It is true that technological evolution requires policy updates, but many changes are only mentioned in passing.

The Procedure

The change to Claude's terms of use and privacy policy will go into effect on 28 September next and by that date users are necessarily asked to choose whether or not to consent to the request. If not, it is necessary to object by opting out. In any case, it is indispensable to reply in order to continue accessing the related tools.

Accessing Claude will open an information window. To refuse, one must click on the bar next to 'Help improve Claude' to set it to off (from black to white), otherwise leave everything unchanged. Those who accept will see the chat data used for training, but Anthropic promises privacy through obfuscation and filtering, with no sharing with third parties. The data will be retained five years instead of the standard 30 days.

The new rules apply to users of Claude Free, Pro and Max, including those using Claude Code. Business customers using Claude Gov, Claude for Work, Claude for Education or API access remain protected, however - a distinction that follows OpenAI's distinction for its enterprise customers.