# Safe Use Policy

## Scope

This policy applies to Spain Reference Personas Frontier and downstream systems that use it for simulation, evaluation, prompt conditioning, benchmark development or service design.

## Allowed Uses

The dataset may be used for:

- controlled LLM evaluation;
- benchmark development;
- prompt-conditioning experiments;
- subgroup analysis at synthetic cohort level;
- service design and user-research prototyping;
- civic AI and public-interest AI research;
- localization evaluation for Spain-facing systems;
- synthetic scenario testing;
- educational and reproducibility work.

## Restricted or Prohibited Uses

Do not use the dataset for:

- surveillance;
- immigration enforcement;
- law-enforcement targeting;
- attempts to identify or approximate real people;
- individual-level persuasion;
- political microtargeting;
- voter manipulation;
- automated eligibility decisions;
- credit, employment, housing, insurance or benefits decisions;
- legal, medical or mental-health advice without qualified human oversight;
- claims about real citizens, real voters, real customers or real patients;
- replacing fieldwork, surveys, administrative microdata or domain-specific validation.

## Synthetic-Persona Interpretation Rule

Synthetic personas are not real people. They are structured simulation artifacts. They must not be described as respondents, citizens, voters, customers, patients or service users unless the word "synthetic" is explicit.

## Political and Civic-Risk Rule

The dataset may be used to test robustness, bias, representation, deliberation and civic-service access. It must not be used to optimize persuasion of real groups, produce manipulative messaging, or target political content at individual or subgroup level.

## Deployment Rule

Any system that uses this dataset in a real user-facing setting should include:

- clear disclosure that outputs are model-generated;
- source citation where relevant;
- human review for sensitive contexts;
- refusal or escalation behavior for legal, medical, violence, trafficking, asylum or emergency issues;
- monitoring for overclaiming and unsafe advice.

## Reporting Concerns

If a downstream user identifies a privacy, safety or misuse concern, they should report it through the repository issue tracker or the contact channel listed in the dataset card.