The race between AI models is heating up, and while GPT-5 may not have been as good as anticipated, that does not mean that other competitors are resting on their laurels either. We already know that Meta"s Superintelligence team is actively working in this space, and now, Anthropic is updating its terms and conditions to allow its Claude assistant to be trained on your chat transcripts.
Anthropic says that it will soon enable Claude Free, Pro, and Max customers to choose if they want Claude to be trained on their chat transcripts. The company"s rationale is that this will allow it to build better models and strengthen its guardrails. These updated terms will not apply to Claude Commercial customers, which include Claude for Work, Claude Gov, Claude for Education, or API-based consumption.
If you"re a new user, you"ll be able to set preferences for data sharing upon sign-up. Meanwhile, existing users will receive a pop-up notification requesting them to make a decision. They can make that decision right now or defer until September 28, at which point they"ll necessarily have to make a choice. It is important to note that this applies only to new chats or resumed sessions, not older chats.
In addition, Anthropic is increasing its data retention period from 30 days to five years if you choose to allow Claude to consume your data for training. The reasoning behind this clause in particular is understandable because models take a long time to train, typically months or even years.
As such, it makes sense to have Claude have access to your data for a longer period of time than usual. Fortunately though, users who initially decide to share chat data can disable the option through Privacy settings. This would also stop Anthropic from using your previously accessible data in future model training, but it still may leverage it in models that are training or have been trained.