| copyright |
|
||
|---|---|---|---|
| lastupdated | 2026-06-03 | ||
| keywords | data, secure, encrypt, cos, bucket, storage | ||
| subcollection | inference |
{{site.data.keyword.attribute-definition-list}}
{: #mng-data}
To ensure that you can securely manage your data when you use {{site.data.keyword.instructlab_short}}, it is important to know exactly what data is stored and encrypted and how you can delete any stored data. {: shortdesc}
{: #data-storage}
Your data is stored differently depending on how you use {{site.data.keyword.instructlab_short}}.
{: #data-ma}
{{site.data.keyword.instructlab_short}} uses an {{site.data.keyword.cos_full}} instance to store your taxonomy and training data. All objects that are stored in {{site.data.keyword.cos_short}} are encrypted by default. For added security, you can use {{site.data.keyword.keymanagementserviceshort}} or {{site.data.keyword.hscrypto}} to encrypt the data in your {{site.data.keyword.cos_short}} bucket as well. For more information, see Encrypting your data in the {{site.data.keyword.cos_short}} documentation.
The knowledge documents that you reference in your qna.yaml files are not stored by {{site.data.keyword.instructlab_short}}.
{: #using-byok}
You must enable {{site.data.keyword.instructlab_short}} to use {{site.data.keyword.cos_short}} by creating a service authorization. If you delete this authorization, your {{site.data.keyword.instructlab_short}} instance loses access to your {{site.data.keyword.cos_short}} bucket. However, any existing data remains in your bucket.
{: #data-inference}
Chat completion responses are stored so you can retrieve them. You have full control over this data and can delete it as needed. No inference data is stored in a {{site.data.keyword.cos_short}} bucket.
{: #data-delete}
All data used or generated by {{site.data.keyword.instructlab_short}} for model alignment is stored in your {{site.data.keyword.cos_short}} bucket and remains there unless you manually delete it. For information on deleting data from COS, see Emptying a bucket.
For inference, you can delete individual chat completions. If you delete a {{site.data.keyword.instructlab_short}} project, all chat completions are deleted as well.
{: #data-jobs}
Once a job is completed, the generated data related to model alignment is uploaded to your {{site.data.keyword.cos_short}} bucket. Deleting a job does not delete the data in the bucket. If you cancel a job before it is completed, any data that was generated before the cancellation is uploaded to your bucket. Canceling a job does not delete the data in the bucket.
For information on the data uploaded to your {{site.data.keyword.cos_short}} bucket, see What's in my {{site.data.keyword.cos_short}} bucket after data generation?.
{: #service-delete}
If you delete your {{site.data.keyword.instructlab_short}} instance, relevant data still exists in your {{site.data.keyword.cos_short}} bucket. Deleting the {{site.data.keyword.instructlab_short}} instance does not delete the data in the {{site.data.keyword.cos_short}} bucket.
{: #data-restore}
Data that is deleted from your {{site.data.keyword.cos_short}} bucket cannot be recovered.
{: #data-access-remove}
Access to {{site.data.keyword.instructlab_short}} service instances for users in your account is controlled by {{site.data.keyword.iamlong}} (IAM). Every user that accesses the {{site.data.keyword.instructlab_short}} service in your account must be assigned an access policy with an IAM role, which determines what actions a user can perform within the context of the service or specific instance that you select. To remove a user's access policy, see Removing access by using the CLI or Removing access in the console.