Anthropic's Claude now trains AI with user chats for up to 5 years
Serge Bulaev
Anthropic has changed its policy so that chats from users on Claude's Free, Pro, or Max plans are now kept for up to five years and may be used to train future AI models by default. Users can opt out, which returns chat storage to 30 days and stops their conversations from being used in training. Some data, like policy violations, might be kept longer, and deleted chats may stay in backup logs for up to a month. Business and API customers are not affected, as their data is not used for public model training. Experts suggest that users on consumer plans should be careful, as their data might be more risky if not properly managed.
Following a major policy update, Anthropic's Claude now retains user chats for up to 5 years when users enable the 'Help improve Claude' setting. According to Anthropic's updated terms, this change affects conversations from Free, Pro, and Max tiers only when users opt-in to model training. This extended data retention represents a significant shift from the previous policy structure.
This opt-in model means that any personal or professional information entered into a consumer account could be stored for years if users choose to enable training. While Business and API customers are exempt from this public model training, individual subscribers now have clearer control over their data privacy through explicit consent.
What This Policy Change Means for Users
Anthropic updated its data policy for Claude's Free, Pro, and Max users, making the 'Help improve Claude' setting available with up to five years of retention for chats used in AI model training. Previously, chats were retained indefinitely by default and used for training unless users opted out. Users must now actively opt-in to training, with non-training chats deleted after 30 days.
Prior to August 2025, standard user chats were retained indefinitely by default and were used for model training unless the user opted out. The 30-day retention was specific to API data or users who disabled chat history. Now, enabling the training setting extends retention to approximately 1,825 days (five years). Keeping the setting disabled maintains the 30-day retention period. Note that even deleted chats may remain in backup systems for up to 30 days, a practice described as a security buffer by Voibe.
How to Opt Out and Protect Your Data
A single toggle in your account settings controls both data retention and training consent. To keep your data private:
- Open Account Settings.
- Navigate to Privacy or Data Sharing.
- Keep the "Help improve Claude" toggle OFF.
- Confirm your choice when prompted.
Anthropic clarifies that this setting applies immediately to all future conversations. However, any data collected from users who previously opted in may still be used for research, a point stressed by ForbiddenAI.
Exceptions to Standard Data Retention
Even if you don't enable training, some data may be retained longer for safety reasons. Policy-flagged conversations are retained for up to 2 years, while trust and safety classification scores are retained for up to 7 years, according to FAQs cited by The Register. These safety protocols operate independently of user privacy settings and can impact compliance.
The Risk for Professional Use on Consumer Plans
Experts emphasize that data governance now hinges on which subscription plan is used. For example, proprietary code pasted into a personal Pro account could potentially be exposed in other users' outputs years later if training is enabled. Research on latent memorization highlights that completely removing data from a trained large language model is a significant technical challenge.
To mitigate this risk, enterprises should use Team or Enterprise plans, which are excluded from public model training. Smaller organizations using consumer tiers should ensure the training setting remains disabled for all users handling sensitive or client information.
Data Policy Quick Reference
| Setting | Retention window | Model training |
|---|---|---|
| Training enabled (Free/Pro/Max) | Up to 5 years | Yes |
| Training disabled (default) | 30 days | No |
| Business tiers | 30 days* | No |
*Business tier retention is for abuse and misuse monitoring only.
The 5-year retention applies only to users who explicitly opt-in to training. Users who do not opt-in have their data deleted after 30 days. Corporate security teams may need to audit Claude consumer accounts used on company devices to ensure training remains disabled for sensitive data handling.
FAQ: Claude's New Data Retention and Training Policies
When did Claude's privacy policy change take effect?
The policy shift became effective in late August 2025, with Anthropic announcing the changes on August 28, 2025. Users were given until October 8, 2025 to opt-in to training. Users who did not opt-in have their data deleted after 30 days rather than being subject to extended retention terms.
Which Claude plans are affected by the 5-year data retention policy?
The policy applies exclusively to consumer tiers: Free, Pro, and Max plans when users enable the 'Help improve Claude' setting. Business tiers - including Team, Enterprise, Gov, and Education plans - remain unchanged and do not use conversations for model training by default. This creates a critical divergence where exposure is governed by subscription plan and user choice rather than automatic enrollment, meaning sensitive work performed on consumer accounts faces fundamentally different privacy protections than identical work on business accounts.
How can I prevent my Claude conversations from being used for AI training?
To maintain 30-day retention and avoid training use, you must keep the "Help improve Claude" setting disabled:
- Navigate to Account Settings
- Locate Privacy Settings or Data Sharing
- Keep "Help improve Claude" toggled OFF
Important limitations:
- The setting applies to all conversations based on your current preference
- The change takes effect immediately for future interactions
- Deleting a conversation removes it from the interface and purges it from back-end storage within 30 days, preventing future training use
What happens to my data if I enable Claude's model training?
If you enable the training setting, your conversations are:
- Retained for up to 5 years (approximately 1,825 days) in de-identified form
- Used to train future Claude models and for safety improvements
- Subject to extended retention even if flagged: trust and safety classification scores may be retained up to 7 years, and policy-flagged conversations up to 2 years
Even without training enabled, some data retention occurs: 30 days for technical and security purposes, with flagged safety content potentially held longer regardless of your settings.
How does Claude's privacy approach compare to other AI chatbots?
Major consumer AI platforms have varying approaches to training inclusion, with significant differences in trade-offs:
| Platform | Default Training | Retention if Training ON | Opt-Out Trade-off |
|---|---|---|---|
| Claude | OFF (opt-in required) | Up to 5 years | No history loss - you can keep chat history while avoiding training |
| ChatGPT | ON (opt-out required) | Indefinite | No history loss - clear opt-out in Settings |
| Gemini | ON (opt-out required) | Up to 18 months (3 years for feedback) | History deletion required - you cannot keep history and stop training simultaneously |
For business users, all three providers offer stronger protections: Claude's Team and Enterprise tiers, ChatGPT's Enterprise tier, and Gemini's Workspace plans provide limited data retention and guarantee that data is not used for public model training.