I've set up an AI solution for my project that automatically gathers logs, analyzes them, and sends out email reports with suggested fixes. However, my tech architect is worried about data security since we're sending application log data to AI. To address this, I've implemented data masking to ensure that sensitive information like IP addresses and host details are not sent, only error details and exceptions are relayed. The AI is still able to provide solutions based on this masked data, which is great.
Now, I'm curious about whether the AI is actually learning from our logs. Is there a way in Azure AI Foundry to ensure that the AI is not learning or storing my log data? I need to confirm that my data remains secure enough for production workloads. Also, if there are any official Microsoft documents stating that AI is not learning from my data, that would be really helpful! For your information, I've configured GPT-4o for my solution in Azure AI Foundry, in case that provides any additional context.
5 Answers
No, they don’t train on your data at all! It’s designed to keep your information secure.
You can find more details in this [document](https://learn.microsoft.com/en-us/azure/foundry-classic/foundry-models/faq). It touches on how your data is handled and ensures no learning occurs from it.
The bigger concern isn’t just whether it's learning—but also if anything sensitive is getting sent and who can access the results later. It sounds like you’re on the right track with the masking approach!
Just to confirm, the Foundry models don’t use company data to retrain the models, and your data is kept private—never shared with model providers. Here’s the [source](https://learn.microsoft.com/en-us/azure/foundry-classic/foundry-models/faq#do-you-use-my-company-data-to-train-any-of-the-models--).
Yes, you can rest easy! Azure Foundry doesn’t train on your data. It’s built to be a secure enterprise platform, so your data privacy is taken seriously. You can check out their official FAQs for more details.
I've handled the masking, but I'm particularly worried about whether the AI can still use or learn from my data.