Cloud Engineer · Step 7 of 8 · 8 min
AI services in the cloud and data privacy
Know what cloud AI platforms offer and what to check before sending them company data.
Models as a service
Each major provider offers access to large language models and other AI through its cloud: Amazon Bedrock on AWS, Microsoft's Azure AI services including Azure OpenAI, and Vertex AI on Google Cloud. You call a model through an API, inside the same account, network and access controls as the rest of your systems.
They also offer building blocks around the model: search over your own documents, agents, evaluation tools and content filters.
Why companies use the cloud route
Using a model through your cloud account usually comes with business terms: your prompts and outputs are not used to train the provider's models, data can stay in a chosen region, and access is controlled with the same IAM as everything else. That is often what legal and security teams need before they approve AI use.
Read the actual terms for the service and region you use. Do not assume; features, regions and data handling differ between models.
What to check before sending data
Ask four questions. Where is the data processed and stored? Is it used for training? Are prompts and outputs logged, by whom, and for how long? Who in my company can call this model and see the logs?
Keep personal and secret data out of prompts unless the service is approved for it. Apply least privilege to AI endpoints just as you would to a database.
Costs and limits
AI calls are usually priced per token, with larger models costing more. A loop or a busy app can spend quickly, so set budget alerts and quotas. Start with the smallest model that does the job well.
Try it
Read the data terms of one AI service
- 1.Pick one cloud AI service from the provider you prefer.
- 2.Find its data privacy or data usage page in the official docs.
- 3.Answer the four questions from this step: where, training, logging, access.
- 4.Ask a chatbot the prompt below and check its answers against the page you found, not the other way round.
I am evaluating [service] for company use. Write a one-page checklist of data privacy and security questions to answer before we send it customer data, grouped by: data location, training use, logging and retention, access control and cost limits.