← Back to Parity

Generative AI Use Policy

Last updated: April 9, 2026

Parity Talk ("Parity," "we," "us," or "our") uses generative artificial intelligence as a core part of the Parity mobile application. This policy explains how AI is used, what data is processed, what limitations exist, and how we ensure responsible use.

1. What AI model does Parity use?

Parity uses Anthropic's Claude language model via their commercial API. Specifically:

• Claude Sonnet is used for all user-facing translations and Talk mode conversations, where quality and nuance are critical
• Claude Haiku is used for background summarization tasks (generating learned patterns from conversations), where speed and cost efficiency are prioritized

We use Anthropic's commercial API tier, which means Anthropic does not use your inputs or outputs to train their models. For details, see Anthropic's privacy policy and Anthropic's usage policy.

2. What data is sent to the AI?

When you use Translate or Talk mode, the following information may be included in the AI prompt:

2.1 Translate mode

• Your neurotype(s) and relevant communication challenges
• Your connection's name, neurotype(s), age, relationship to you, and description
• Learned patterns from previous conversations with that connection (AI-generated summaries, not raw conversation history)
• The specific text you entered for translation
• The direction of translation (to them or from them)

2.2 Talk mode

• Your neurotype(s) and relevant communication challenges
• If a connection is selected: their name, neurotype(s), relationship, and learned patterns
• The conversation messages within the current Talk session

2.3 Background summarization

• The input and output of a completed translation
• Existing learned patterns for the relevant connection
• The connection's name (for context)

2.4 What is NOT sent

• Your full conversation history is NOT sent with every request
• Your email address or Apple ID is NOT sent
• Other users' data is NOT included in your prompts
• Your payment or subscription information is NOT sent

3. How is AI-generated content used?

• Translations: AI generates suggested rephrased communication based on the recipient's neurotype and known communication patterns. These are presented as suggestions for you to review, edit, and optionally use.
• Talk responses: AI generates conversational coaching responses informed by your profile and connection context. These are informational and supportive, not therapeutic.
• Learned patterns: After each translation, AI generates a brief summary of observed communication patterns. These summaries are stored in your connection profile and used to improve future translations. You can view and edit these summaries at any time.
• "Why this works" explanations: AI generates educational explanations of why a particular translation approach is suited to the recipient's neurotype. These are general communication insights, not clinical assessments.

4. Limitations of AI-generated content

You acknowledge and agree that:

• AI can produce incorrect or inappropriate responses. No AI model is perfect. Translations may miss cultural context, relationship dynamics, or situational nuances that a human would catch.
• AI does not understand your relationship. The AI constructs an approximation of your communication dynamic based on the information you provide. It cannot observe body language, tone of voice, facial expressions, or environmental context.
• AI-generated patterns may be wrong. The learned patterns feature uses AI to infer communication preferences from limited data. These inferences may be inaccurate. You should review and correct them regularly.
• AI is not a substitute for professional support. The AI cannot diagnose conditions, provide therapy, manage crises, or replace any form of professional medical, psychological, or communication intervention.
• Responses vary. The same input may produce different outputs at different times. AI outputs are probabilistic, not deterministic.
• AI reflects training data limitations. The AI model's understanding of neurodivergent communication is based on its training data, which may not fully represent all neurotypes, cultural contexts, or individual experiences.

5. Data retention and AI training

• Anthropic does not retain your data. Under Anthropic's commercial API terms, inputs and outputs are not stored beyond the time needed to generate a response and are not used for model training.
• We do not train AI models. Parity does not train, fine-tune, or otherwise modify AI models using your data. We use Anthropic's pre-trained models via their API.
• We do not use your data for AI training. Your personal data, conversation content, and connection information are never used to train any AI model, whether ours or a third party's.
• Learned patterns are stored locally. AI-generated summaries of communication patterns are stored on your device (and in your encrypted cloud backup if enabled). They are not stored on our servers or shared with Anthropic outside of future API calls for that connection.

6. Content moderation

The AI model includes built-in safety measures provided by Anthropic. Additionally, our system prompts instruct the AI to:

• Never provide medical diagnoses or treatment recommendations
• Never provide crisis intervention or suicide prevention guidance
• Always frame responses as communication suggestions, not directives
• Decline requests that could facilitate manipulation, abuse, or harm
• Acknowledge uncertainty rather than fabricating confident responses
• Respect the validity of all neurotypes without pathologizing

Despite these measures, no content moderation system is perfect. If you encounter AI-generated content that is harmful, inaccurate, or inappropriate, please report it to [email protected].

7. Your responsibilities

• Review before using. Always review AI-generated translations before using them in actual conversations. Use your own judgment about what is appropriate for your specific situation.
• Edit learned patterns. Regularly review the "What Parity has learned" section for each connection and correct any inaccurate patterns.
• Provide accurate context. The quality of AI responses depends directly on the accuracy and completeness of the information you provide in your profile and connection profiles.
• Do not rely solely on AI. Parity is a tool to assist your communication, not to replace your judgment, empathy, or understanding of the people in your life.
• Do not upload others' private information without consent. When creating connection profiles, ensure you have appropriate consent or authority to share information about that person.

8. Changes to this policy

We may update this Generative AI Use Policy as our AI practices evolve or as regulatory requirements change. We will notify you of material changes by posting the updated policy and updating the "Last updated" date. Continued use of the Service constitutes acceptance of the updated policy.

9. Contact

For questions about our AI practices:

Email: