AIQu VEIL for Snowflake

AIQu VEIL encodes sensitive feature columns into latent vectors. Masking policies enforce this at query time. Analysts see encoded vectors. Admins see raw data. No data leaves your Snowflake account.

Use cases

Protect sensitive features for ML. Prices, dimensions, behavioral signals are encoded into vectors that preserve statistical properties but can’t be reversed to raw values.

Share data without exposing raw features. Encoded views let external parties or cross-functional teams work with protected data.

Replace manual anonymization. Apply a masking policy once. Every query through the view is encoded automatically.

How it works

Train an autoencoder on your data. VEIL creates a secure view with a masking policy on the features column. Non-privileged roles get encoded vectors. Admins get raw values. Training runs inside your account on CPU or GPU.

Roles

Role	Access
app_admin	Train encoders, create views, manage services, schedule retraining
app_user	Query views, call encode function

app_admin inherits app_user.

GRANT APPLICATION ROLE <app_name>.app_admin TO ROLE my_admin_role;
GRANT APPLICATION ROLE <app_name>.app_user TO ROLE my_analyst_role;

Compute

Service	Default	Options
Encoding	CPU_X64_XS	Scales to 3 replicas. Larger instance families available.
Training	CPU	GPU optional. One job at a time.

All compute is provisioned by the app.

Privileges

Five privileges requested through Snowsight at install. Services start once all are granted.

Privilege	Purpose
CREATE COMPUTE POOL	CPU and GPU pools
BIND SERVICE ENDPOINT	Route queries to encoding service
CREATE WAREHOUSE	Read training data
EXECUTE TASK	Scheduled retraining and billing
EXECUTE MANAGED TASK	Serverless scheduled tasks

Masking policies require no additional privilege. The app owns the policies and the views.