Salesforce Agentforce
Salesforce says Agentforce runs through the Einstein Trust Layer with zero data retention agreements for third-party LLMs, and Salesforce does not currently use customer data to train generative AI models without an express opt-in.
Third-party LLM providers are stated not to retain customer data or use it for training. Salesforce says any future generative-model training on customer data would require product-term opt-in.
No opt-out is required for the stated generative-AI baseline. Predictive Einstein global-model features are a separate path and may have their own opt-out controls.
Agentforce can ground responses in CRM and connected data through RAG, masking, and permission controls; the training answer does not remove connector, action, and oversharing risk.
Salesforce describes zero data retention with current third-party LLM providers for Agentforce processing; underlying CRM records remain subject to Salesforce service retention and customer configuration.
Salesforce recommends human review of model responses before consequential use; the reviewed Trust Layer materials focus on controls rather than one support-review disclosure.
Foundational reference pages for teams adopting, deploying, and operating production AI with inspectable controls.