TL;DR
- Verify whether model-improvement training is opt-in or default for Anthropic API.
- Confirm retention for deleted chats or logs and whether data can be used for training.
- Ensure your access path is officially supported for this plan.
Scope
This page covers terms you should verify for Anthropic API. It is a snapshot checklist, not legal advice.
How to use this page
- Confirm you are using the Anthropic API (not Claude.ai or Claude Code).
- Work through the checklist and capture the current policy date.
- Re-check monthly or after any policy update.
- Return to the Claude terms hub for plan comparisons.
What changed recently
See /posts/anthropic-tos-changes-2025/ for the 2025-2026 timeline and enforcement shifts.
Terms snapshot checklist (as of Feb 2026)
Training use: API usage is opt-in for training—default is OFF. Must explicitly enable “model improvement” in API settings to allow training use. This is the inverse of consumer plans.
Retention:
- Standard API: 30 days for request logs (unless training enabled)
- If training enabled: Up to 5 years retention in de-identified form
- Commercial terms available with custom retention and deletion schedules
Third-party access:
- Fully supported: Any client using your own API key is legitimate
- Use your own key: Tools like OpenCode self-hosted mode, Cursor, custom scripts—when configured with YOUR API key—are officially supported
- NOT supported: Tools spoofing consumer OAuth tokens or pretending to be official clients
- API is the safest path for third-party integrations
Data export/deletion: Full audit logs available; commercial terms include custom data handling agreements.
Enterprise/DPA:
- Standard: Available via self-serve in API dashboard
- Custom: Enterprise agreements include negotiated DPA and BAA (for HIPAA)
- Commercial Terms: Section D.4 prohibits using API to build competing products or train competing models
Evidence / signals
- Official Terms of Service and Privacy Policy.
- API documentation for supported access paths and usage.
- Account settings that show model-improvement toggles.
How to verify
- Read the current Privacy Policy and Terms of Service.
- Check account settings for model-improvement toggles.
- Confirm official client lists and API documentation.
If you already shared data
- Review current settings and disable training if needed.
- Rotate keys or revoke access for unofficial tools.
- Document what data was shared and when.
What would invalidate this
Any new policy effective date, API access change, or plan-level terms update should replace this checklist.
Terms deltas (auto)
Verified deltas vs baseline
No verified deltas yet.
Potential deltas to verify
- Training use: Not confirmed; verify API data usage and training policy. (baseline: Consumer plans: model-improvement setting controls training (verify current policy).)
- Retention (deleted data): Not confirmed; verify API data retention and deletion behavior. (baseline: Consumer plans: deleted chats removed within ~30 days when training is disabled (verify).)
- Retention (training data): Not confirmed; verify API training retention window. (baseline: Consumer plans: training data may be retained up to 5 years if enabled (verify).)
- Third-party access: Use documented API access only; verify allowed client patterns. (baseline: Official clients only; unofficial access may be blocked (verify).)
- Official clients: API keys + official docs; verify allowed integrations. (baseline: Claude app/web, Claude Code, Anthropic API (verify current list).)
Terms snapshot
| Field | Value | Status | Source |
|---|---|---|---|
| Training use | Not confirmed; verify API data usage and training policy. | unknown | |
| Retention (deleted data) | Not confirmed; verify API data retention and deletion behavior. | unknown | |
| Retention (training data) | Not confirmed; verify API training retention window. | unknown | |
| Third-party access | Use documented API access only; verify allowed client patterns. | needs_verification | /verify/anthropic-api-clients/ |
| Official clients | API keys + official docs; verify allowed integrations. | needs_verification | /verify/anthropic-api-clients/ |
| Enterprise DPA | Ask sales/support if you require a DPA; availability unconfirmed. | unknown |
Related links
- Verification methodology
- Claude terms overview
- Anthropic policy claims
- Anthropic API client analysis
- Third-party access risk
- Anthropic TOS changes timeline
- /posts/anthropic-tos-changes-2025/
- /compare/claude-vs-openai/pricing/
- /value/smart-spend/