Companies that run Anthropic's newest model, Claude Fable 5, through Amazon Web Services' Bedrock platform face a notable condition: using it means sharing inference data back with Anthropic.
That is the central claim in a report from InfoQ, surfaced via Google News. "Inference data" refers to the inputs and outputs generated when an AI model is actually put to work answering prompts — the queries users send and the responses the model returns.
The detail matters because Bedrock is Amazon's managed service for accessing third-party AI models, and many businesses choose it precisely for the enterprise controls and data-handling assurances a major cloud provider offers. A requirement to route inference data to the model's maker cuts against the expectation, common among Bedrock customers, that their data stays within their own cloud environment.
According to InfoQ, the data-sharing arrangement is specifically tied to running Fable 5 on Bedrock, signaling that the terms attached to a model can vary by where and how it is deployed.
Why it matters: as organizations weigh which AI models to adopt, the fine print on data sharing — not just price or performance — is becoming a decisive factor in whether a model is safe to use with sensitive information.