Anthropic updated its privacy policy this week with a clause worth reading closely before the changes take effect next month. The specific shift: the previous policy committed to protecting user data except when required to do so by a court or legal process. The updated version protects user data except when Anthropic decides otherwise.
The difference sounds subtle but isn't. "Unless a court orders it" is an objective, external standard - a specific legal threshold that must be met before disclosure happens. "Unless we decide otherwise" puts that decision inside the company, with no public standard attached to it.
What This Means in Practice
For most people using Claude for everyday tasks - drafting emails, summarizing documents, generating ideas - the practical risk is probably low. The change doesn't mean Anthropic is planning to sell your data or hand it to third parties casually.
For anyone using Claude for sensitive work - legal research, medical consultations, confidential client strategy, competitive analysis - the threshold change matters more. The previous language gave users something closer to a contractual standard tied to external legal requirements. The new language gives users a stated preference that the company can override at its own discretion.
What to Check
Users with privacy concerns should review Anthropic's updated policy directly before it takes effect. A few things worth verifying: the opt-out setting for having conversations used in model training (a separate provision, but worth confirming is off for sensitive accounts), and whether your use case involves data governed by specific regulations like HIPAA or GDPR, which may create additional obligations regardless of what Anthropic's policy says.
Enterprise and API customers should verify whether their data processing agreements carry different terms - enterprise contracts often include stronger protections than the consumer policy.