Agent Development Security Compliance Guide: Children's Mode

Functional overview

Children’s mode (also known as child safety mode) is a compliance and safety capability designed for children’s AI products and agents. When a product is intended for children or needs to comply with children-related regulations, the platform enforces restrictions on capabilities and data usage through children’s mode. This aims to align with regulatory requirements such as the US COPPA, EU AI Act, and EU GDPR, mitigating compliance risks and ensuring safe usage for children.

Applicable scenarios and value

Standard mode vs. children’s mode

For your quick reference and configuration, the table below outlines the differences between standard mode and children’s mode by feature, including whether a function can be toggled on in children’s mode (that is, user/developer can enable or disable it).

Data and commercialization commitment (platform)

• Children’s data not used for model training: Without explicit compliance and authorization, children’s data will not be used for model training.
• Children’s data not used for advertising: Children’s data will not be used for ad targeting or similar commercial purposes.

Configure and use "child safety mode" for your product

Solution: Default marking for "dedicated children’s AI solution"

• Once the solution is marked, products created based on this solution must meet child compliance requirements. Examples: mandatory protocol enforcement, disabling features that record children’s information, and ensuring associated agents operate in child safety mode.
• In the protocol configuration, children’s privacy protocols must be enabled and set to mandatory consent as per product requirements.

Agent: Select child safety mode

• When creating or editing an agent, you must select its operating mode as standard mode or child safety mode.
• If a solution is marked as dedicated for children, any agent associated with products under this solution must have child safety mode enabled. If it is not enabled, the binding process will prompt you to enable it on the agent configuration page. Otherwise, binding cannot be completed.
• The agent list will display a child safety mode or standard mode label for easy identification and management.
• After child safety mode is selected, constraints like "mandatory", "off by default", and "regionally restricted" mentioned above will take effect automatically. Some capabilities may appear grayed out in the UI with a note explaining they cannot be enabled due to child safety mode.

Skills and workflows

• Skills: Some skills have special requirements in children’s mode, for example:
  • Music/children’s songs and stories can only use the children’s version or compliant sub-version.
  • Device control skills are limited to self-control only.
  • Emotion-related skills are unavailable in some regions.
• Workflows: When an agent carries a child label, large language model nodes in the workflow will be injected with system prompts related to child compliance. If nodes or tools unsuitable for children are used, prompts will be provided during stages such as trial runs.

FAQs

Q: What are the differences between children’s mode and standard mode?

Refer to the earlier section Feature list: Standard mode vs. children’s mode. This table lists the differences by feature under both modes, including whether functions can be toggled in children’s mode (mandatory, off by default, and regionally restricted), allowing for a clear comparison at a glance.

Q: If a solution is marked for children, can it be associated with an agent that does not have child safety mode enabled?

No. When associating an agent, the system will verify whether child safety mode is enabled. If not, you must first enable it in the agent configuration before completing the binding.

Q: For products using a child-dedicated solution, can developers disable "parental verification" or "children’s privacy protocol"?

No. Parental verification and children’s privacy protocol are mandatory and cannot be disabled in children’s mode to ensure compliance and safety.

Q: Is emotion recognition completely unavailable in children’s mode?

The platform does not record children’s emotional data, whereas capabilities like emotion feedback can be retained. In regions like the EU, emotion recognition capabilities intended for children are disabled. Other regions are subject to the product and local regulations.

Q: Why is voice cloning/voiceprint unavailable in the US?

Local regulations classify voiceprints as sensitive personal data and restrict access to such data from specific regions. Therefore, in relevant regions like the US, children’s products will have voiceprint recognition and voice cloning functions disabled.

Q: When will parental controls (e.g., time limits, dangerous topic warnings) be available?

These features are planned for future versions. The specific details will be confirmed in product release announcements.

Revision and support

• This documentation serves as an external functional and compliance explanation. Specific interactions, permissions, and support scope are subject to the current version of the Tuya Developer Platform.
• Compliance requirements might change due to regional and regulatory updates. Please stay tuned for platform announcements and the latest documentation.
• For inquiries, please contact us via the Developer Platform ticket system, Documentation Center, or your designated platform support channel.
• For updates, please refer to the latest official documentation.