Meta coding-harness subscriptions default to a data-use tier that permits model training

Meta Model API terms distinguish Contributor Services, where submitted Content can be used to train and evaluate models, from Standard Services, where the terms say Content is not used to train Meta Models. Coding Harness Subscriptions start in Contributor status regardless of model selection until the account setting is changed. The switch operates prospectively, so previously submitted materials are not retroactively removed from Contributor rights.

The service terms explicitly prohibit sending sensitive, confidential or personal information to Contributor Services and warn that reused Standard-origin material becomes Contributor material when included in a Contributor request. This is a contractual data-handling boundary rather than proof that a particular customer repository has already been trained into a model. Agent developers must treat prompt and tool context as potential Content under the selected setting.

A user selecting a particular model does not override the Coding Harness Subscription account setting. Under Section 3.3 the subscription is Contributor by default; Section 6 permits reuse of submitted content for model improvement and bars sensitive or confidential input, while Section 5 excludes Standard request content from Meta Model training. Content originating in Standard can nonetheless become Contributor content if reproduced in a later Contributor request. For coding agents the practical boundary is every tool result and file excerpt forwarded as request context, not only the human-written prompt.