Data Handling in Generative AI
Learn how OCI Generative AI handles user data.
- Does OCI Generative AI retain customer-provided training data used to fine-tune a custom model?
-
No. The customer stores and manages their training data in their own customer tenancy (commonly inside an OCI Object Storage bucket). OCI Generative AI's fine-tuning job uses this training data to train a custom model for customer. OCI Generative AI doesn't retain this training data beyond the duration of this training job. This training data is solely used for building a custom model for this customer. The custom model is also a resource that's managed by the customer. This training data is not used to improve the general use cases for OCI Generative AI.
- Does OCI Generative AI retain customer-provided prompts and inputs used for inferencing, on the large language models (LLMs)?
- No, OCI Generative AI doesn't retain customer inputs. A user's input on an inference call is sent to the LLM and a response is generated by the LLM that's returned to the user. Both input and output are not stored inside OCI Generative AI.
- Does OCI Generative AI share prompts and responses, fine-tuning training data, or fine-tuned custom models with third-party model providers such as Cohere or Meta?
- No.
- Is the training data encrypted?
-
The data is encrypted both at rest and in transit. The training data is deleted from the fine-tuning cluster as soon as the training job is completed.
Training data for fine-tuning a model is always double encrypted, by default, by Oracle-managed
AES-256encryption and optionally by customer-managed private keys through OCI Vault service. Customers can delete the data at any time.Oracle encrypts all the data in motion with
TLS 1.2.