Instinct’s privacy policy, last revised August 26, allows information from use of the service to support model evaluation, fine-tuning and training, with an opt-out in settings. That control applies prospectively. The policy preserves an exception for information flagged for safety review and says previously trained models may continue to be used. It separately excludes information received directly through Google Workspace from AI model training and improvement.
Two additional distinctions matter when connecting personal accounts. The terms say materials placed in Vault are used to provide the service rather than train models. Meanwhile, disconnecting an integration does not automatically erase information already collected or indexed from it; the policy provides separate deletion routes. These are descriptions of existing published rules, not evidence of a newly introduced privacy change or a finding that data has been mishandled.
Editorial take: Treat connection, retention and training as three separate decisions. Turning off future access may leave an earlier copy behind, while a training opt-out has a different purpose from deletion. Reviewing those controls before adding sensitive accounts is more useful than assuming one switch governs everything.