Anthropic Seeks Permission to Train AI Model on User Conversations, Updates Privacy Policy

Anthropic announced an update to its privacy policy on Thursday, highlighting that it will now train its existing and future artificial intelligence (AI) models on users' conversations. The San Francisco-based AI company highlighted that it is now giving users an option to either allow or deny permission to train on its chats. This revised privacy policy only affects end consumers and is not aimed at the AI firm's enterprise customers. In addition to the revised privacy policy, Anthropic is now also extending its data retention policy to store users' conversations for a period of five years.

Anthropic's New Privacy Policy Goes Into Effect on September 28

Training AI models on user data is not a new practice among AI companies in Silicon Valley. OpenAI, Google, and Meta have already created provisions where users can consent to sharing their conversations with the companies to train their AI models. So far, Anthropic was an exception to this, as it did not even train its models on the data of those on the free tier.

Anthropic explains that it will begin training its models on users' chats. For this, Anthropic is giving users the choice to consent to data sharing via a new pop-up box that appears when you open Claude's website interface or app. Users have until September 28 to make their decision, as after that, they will have to manually navigate the settings to make the changes.

The company highlighted that the revised policy is only aimed at Claude's Free, Pro, and Max tiers. If a user consents to sharing their data, they will also be sharing the transcripts of when they use Claude Code. Notably, it does not apply to its enterprise services, including Claude for Work, Claude Gov, Claude for Education, or API use. Additionally, it will also not affect those accessing Claude models via third-party platforms.

For both AI training and data retention, only the new chats and chats resumed after accepting the new policy will be used. Old conversations will not be impacted by the revised policy. In case a user deletes a conversation, that will also not be used for future model training. Additionally, Anthropic is also extending its data retention to five years. This only applies to those consenting to share their data; otherwise, the existing 30-day data retention rule will apply.