Dynatrace Intelligence agentic and generative AI data privacy and security
- Latest Dynatrace
- Explanation
- 3-min read
At Dynatrace, we take our responsibility to safeguard your data seriously. Understand how Dynatrace Intelligence agentic and generative AI uses your data and understand your responsibility to keep your data secure.
Prompt data
Although we mask Personally Identifiable Information (PII), we still recommend exercising caution when including personal or confidential information in your prompts.
Agentic Dynatrace Assist doesn't provide any PII masking. In order to protect your data, when Dynatrace Assist detects PII in the user prompt, the request is automatically blocked and the prompt isn't sent to the LLM for processing.
Your prompts are sent to LLMs hosted by enterprise vendors such as Microsoft Azure AI and AWS Bedrock, which power Dynatrace Intelligence agentic and generative AI. Enterprise vendors don't store the data you submit or the responses you receive. The prompts you submit and the responses you receive are used only to serve your experience. Enterprise vendors also don't use the prompts to fine-tune or improve any models or services, or to train models across customers or environments.
For the custom DQL generation model, we use internally generated data that's sourced from the DQL queries executed by Dynatrace employees. The training data doesn't include any identifiable employee information or customer data. We don't use customer data at any stage of the fine-tuning process. To learn more about DQL generation model, see Dynatrace Intelligence DQL generation model.
Each data request is sent to the LLM individually, over an SSL-encrypted service, processed by respective enterprise vendors, and sent back to Dynatrace. If your environment is located in EMEA, your prompts are processed in an EU region. If your environment is located in NORAM, LATAM, or APAC, your prompts are processed in a US region.
Dynatrace may store the prompts submitted to Dynatrace Intelligence agentic and generative AI and the responses provided by the LLMs to understand the use cases, contextualize the feedback on the responses, and identify additional user expectations. If you're using Agentic Dynatrace Assist, Dynatrace may also store results of intermediate tool calls.
Learn more about the Dynatrace Intelligence agentic and generative AI architecture and data flow.
PII masking and blocking
To keep your information secure, Dynatrace provides masking and blocking of Personally Identifiable Information (PII) during your interactions with Dynatrace Intelligence.
PII masking for generative AI
Dynatrace version 1.305+
PII masking is in place for user prompts interacting with all standard generative AI functionality. This ensures that sensitive information included in your prompts won't be forwarded to LLMs hosted by enterprise vendors.
Dynatrace masks the following PII during your interactions with generative AI:
- Email address
- Phone number
- IBAN information
- Credit card number
- IP address
- US bank number
- US social security number
- US ABA routing numbers
- URL query parameters (only parameters with more than two characters are considered)
- Canadian Social Insurance Number (SIN)
In our logs and calls to LLM models, we replace values from the identified patterns above with fake patterns. This means that you'll be able see IBANs in logs, for example, but they'll be made up of random numbers, replacing the original values included in your prompts.
PII blocking for agentic AI
To ensure that sensitive information included in your prompts won't be forwarded to LLMs hosted by enterprise vendors, Dynatrace offers PII blocking for all agentic AI user interaction.
Starting with Dynatrace version 1.345+, PII blocking is disabled for new environments by default. For existing environments, PII blocking remains enabled by default.
PII blocking is designed to protect sensitive data in your prompts, but overly broad detection can degrade agentic AI response quality by blocking false-positives. The ability to select which patterns get blocked allows you to minimize false positives in your environment while keeping the important data protected. We recommend masking your data on ingest to minimize the need to rely on PII blocking that can affect the user experience. In cases where it's not possible, we recommend selectively disabling the patterns that you know are unlikely to appear in your data.
You can control which PII patterns get blocked and manage your PII blocking preferences through 
Settings. To configure the PII blocking in your environment
-
Go to
Settings > Dynatrace Intelligence > Generative and agentic AI. -
Turn on Enable PII blocking.
-
In the Blocked PII patterns section, turn on the PII blocking for the patterns you want to match. You can choose to block interactions with agentic AI where the following PII is detected:
- Canadian Social Insurance Number (SIN)
- Credit card number
- Email address
- IBAN information
- IP address
- Phone number
- URL query parameters (only parameters with more than two characters are considered)
- US bank number
- US social security number