The Audience Agent reasons over your workspace’s data to recommend and build audiences. This page explains what data the agent reads, what it shows you, how it handles personal data, how that data is secured, and where it is processed, so you can confirm the agent fits your privacy and governance requirements before using it.
\nThe agent runs under your authenticated mParticle session and is scoped to the workspaces and permissions your account can access.
\nBy default:
\nMost audience building uses catalog metadata, such as event and attribute names, and aggregate analytics, such as counts, volumes, and coverage. When your permissions and workspace configuration allow it, the agent may also inspect user attributes or event values to help build accurate audience rules. For example, it may confirm whether a subscription status field stores values like “active,” “true,” or “premium” so it can propose a rule that matches your data.
\nThis user and event data may include personal information if your organization stores personal information in those fields. The agent is designed to have access to only the minimum amount of data needed to perform the task, including personal information. If value inspection is restricted, the agent won’t access those stored values and may ask you to provide the exact value needed.
\nWhen the agent uses event and user data, that data is processed under the same customer agreements that govern the rest of the mParticle platform, including processing by the model provider described in Where data is processed.
\nThe Audience Agent is designed not to expose personally identifiable information (PII) in your conversation. It won’t list stored email addresses or phone numbers, it won’t list audience members, and it won’t allow you to extract PII. It can use those fields structurally, for example targeting users who have an email address on file, without revealing anyone’s actual address.
\nThe Audience Agent can show you non-PII business labels. It can show you fields such as a subscription status field storing “active” and “churned,” when seeing the choices helps you pick the right rule.
\nIndividual user data stays permission-gated: the agent reads a specific user’s profile or event history only when your own permissions allow it and you provide that user’s mParticle ID.
\nIf a naming convention in your workspace embeds personal information in a name or label, the agent can read it the same way other platform features can. As a best practice, don’t include personal information, such as a person’s name or email address, in event names, attribute names, or other labels.
\nThe agent is read-only until you explicitly save a new audience definition. It cannot modify existing audiences or workspace configuration.
\nThe agent cannot:
\nCreating an audience and activating it are deliberately separate steps. After you save a definition, you connect and activate it yourself through the normal flow. See Connect an Audience.
\nThe agent may use permission-gated event and user and event data when available and needed (limited by your own user permissions), but most audience work uses metadata, aggregate analytics, and audience definitions. A few additional safeguards govern how it handles your data:
\nThe agent is not a path for exporting personal data or bypassing your workspace’s privacy controls.
\nThe agent uses the same authentication and permission controls as the rest of the mParticle platform, and those controls are enforced by the platform itself, not by instructions given to the agent.
\nData the agent works with is protected the same way as data anywhere else in mParticle: encrypted in transit and encrypted at rest. This includes requests sent to the model provider described in Where data is processed, which are encrypted in transit like all other mParticle traffic.
\nAgent activity is logged under the same security model as other mParticle platform logs, and log access is restricted to a limited set of authorized personnel.
\nThe Audience Agent is an optional supplement to the manual Audience Builder, not a required replacement for it. Both remain available, so you can use whichever you prefer for any given audience. If your organization would rather the agent not be available at all in your account, contact your mParticle account representative to disable it.
\nThe Audience Agent uses a large language model from OpenAI to interpret your requests and generate its responses. When you chat with the agent, your messages and the workspace context needed to answer them, such as catalog metadata and attribute values, are encrypted and sent to OpenAI for processing under mParticle’s agreement with OpenAI.
\nBy default, the agent processes data in the United States using mParticle’s OpenAI account. Your prompts and data are not used to train or improve OpenAI’s models.
\nmParticle may review agent conversations to improve the product, for example to understand where the agent made a poor recommendation or misread a request. mParticle doesn’t use your proprietary data to improve the agent for other customers.
\nIf your organization is based outside the United States or is subject to regional data-processing requirements, confirm that US-based processing meets your requirements before using the agent. mParticle plans to support region-specific processing, beginning with the European Union.
\nThe Audience Agent doesn’t change how data subject requests are handled. It doesn’t create a separate store of consumer data, so requests such as access or erasure continue through mParticle’s existing data subject request process. For more information, see Data Subject Requests.
","frontmatter":{"title":"Data and Privacy","order":3,"seoTitle":null,"seoDescription":null},"parent":{"__typename":"File","modifiedTime":"June 15, 2026"}}},"pageContext":{"slug":"/guides/segmentation/audiences/audience-agent/data-and-privacy/"}},"staticQueryHashes":["2950460867"]}