Data Security is of the utmost priority for Canopy, below you will find information regarding how Canopy does, and does not use data gathered to train tools like artificial intelligence to improve your work and offer greater value within your Canopy instance.
Table of Contents:
Model Training & Testing
Using your data to train our models is Opt-In Required: Your data is never used for training unless you explicitly opt in.
If you choose to Opt-in, you can choose either that your:
Data can be used for testing only
Data can be used for both training and testing
Model Providers:
Canopy builds its own models.
Canopy also uses third-party vendors, both OpenAI & Anthropic, but:
- Canopy data is not used to train general-purpose models by these vendors.
- If opted-in, your data may be used for fine-tuning --only to improve Canopy specific models.
Fine-tuned models and their training data are never shared with others.
How Your Data Is Used
Purpose: If Canopy uses your data, it is used by AI models to generate helpful responses that save time or improve accuracy — replacing tasks that typically require full human interaction.
Execution: These models operate within the same security and permissioning system as normal Canopy users.
Data Storage: Inputs and outputs are stored like any other user action data.
Review:
Input/output pairs may be reviewed to improve performance and ensure safety.
Only authorized Canopy engineers can access this data for support and review.
Unsafe or harmful interactions reported by users are prioritized for review.
Data Security Standards
All data follows Canopy’s privacy and usage policies.
If you opt in to Training Data, there are some additional considerations for your review - There’s a theoretical risk of data being reverse-engineered; however, we have taken the following measures to safeguard against that risk:
Mitigation Measures:
Personally identifiable information (PII) is removed using industry-standard techniques, unless essential (e.g., detecting a social security number).
Models are trained on the minimum necessary data required for the task.
Most models are custom-built per client — reducing cross-client exposure.
For shared models:
- Inputs and outputs follow strict schema and security rules.
- These rules are updated as research and standards evolve.
Need help? Contact Support or ask Penny, our AI Support Bot for assistance.