Covered AI Services

Per Anthropic’s documentation, all customer data and content processed by Fable 5 will be retained for 30 days for trust and safety purposes. If any data or content is flagged by a classifier for safety investigation or Anthropic is legally required to retain it, such data or content may be retained beyond 30 days.

Anthropic requires extended data retention in connection with Fable 5 because some attacks or malicious misuse of the model only becomes visible across multiple requests. Anthropic can more precisely detect larger patterns of misuse when their safeguard classifiers can “zoom out” across many requests. Detecting these threats requires temporarily retaining prompts and outputs so they can be analyzed together, rather than one at a time.

No employee at Anthropic has default access to customer data and content; only automated safety systems process it to identify potential violations of their acceptable use policy.

Per Anthropic’s documentation, human reviewers may access customer data and content only if it is flagged for potential serious harm or upon a customer’s written request. These reviews can only be performed by a small set of approved reviewers through tooling that prevents export, copying, or downloading. Anthropic records every instance of access in a tamper-proof log that reviewers cannot suppress or modify.

Retained data is used for detecting and preventing harm. Retained data is not used to train AI models, and it is not used by Anthropic for any non-safety purpose in connection with any use of Harvey.

Please note that Anthropic may use the “true” or “false positive” labels attributed to flagged outputs to improve their safety systems. However, retained data is never used to train Claude, and any access for a non-safety purpose requires prior approval.

Per Anthropic’s documentation, all customer data and content processed by Fable 5.1 will be retained for 30 days for trust and safety purposes. If any data or content is flagged by a classifier for safety investigation or Anthropic is legally required to retain it, such data or content may be retained beyond 30 days.

Anthropic requires extended data retention in connection with Fable 5.1 because some attacks or malicious misuse of the model only becomes visible across multiple requests. Anthropic can more precisely detect larger patterns of misuse when their safeguard classifiers can “zoom out” across many requests. Detecting these threats requires temporarily retaining prompts and outputs so they can be analyzed together, rather than one at a time.

No employee at Anthropic has default access to customer data and content; only automated safety systems process it to identify potential violations of their acceptable use policy.

Per Anthropic’s documentation, human reviewers may access customer data and content only if it is flagged for potential serious harm or upon a customer’s written request. These reviews can only be performed by a small set of approved reviewers through tooling that prevents export, copying, or downloading. Anthropic records every instance of access in a tamper-proof log that reviewers cannot suppress or modify.

Retained data is used for detecting and preventing harm. Retained data is not used to train AI models, and it is not used by Anthropic for any non-safety purpose in connection with any use of Harvey.

Please note that Anthropic may use the “true” or “false positive” labels attributed to flagged outputs to improve their safety systems. However, retained data is never used to train Claude, and any access for a non-safety purpose requires prior approval.