Tarosyn Legal

AI Safety Policy

Last updated May 31, 2026

Tarosyn LLC ("Tarosyn," "we," "us," or "our") uses artificial intelligence to power tarot readings, horoscopes, dream interpretations, mystic personas, and other spiritual insight features. We believe AI can be a meaningful tool for reflection and self-discovery — and that this potential comes with a responsibility to deploy it carefully. This policy describes how we limit the risks of AI-generated content, what uses we prohibit, and the safeguards built into our platform.

1. disallowed uses of ai

Tarosyn's AI features are designed for spiritual exploration, reflection, and personal insight. The following uses are expressly prohibited — both by us in system design and by our users under the Terms of Service:

Medical diagnosis or treatment advice. AI readings must not be used as a substitute for professional medical care, diagnosis, or treatment recommendations. Legal advice. AI-generated content does not constitute legal counsel. Users must consult a qualified attorney for legal matters. Financial advice. Readings, horoscopes, and mystic messages are not investment guidance, financial planning, or trading recommendations. Psychological or crisis counseling. Our AI is not a licensed therapist or crisis-line service. Users in acute distress should contact a qualified mental health professional or emergency services. Generating harmful, hateful, or abusive content. Users may not craft prompts intended to produce content that is violent, sexually exploitative, harassing, defamatory, or discriminatory. Targeting or manipulating other people. AI features must not be used to build harmful profiles of, harass, stalk, or psychologically manipulate other individuals. Deceptive impersonation. Passing off AI-generated content as authored by a specific real person (including celebrities or other users) in a misleading way is prohibited. Circumventing platform safeguards. Attempting to jailbreak, prompt-inject, or otherwise bypass our content filters or model instructions is a violation of our Acceptable Use Policy.

Violations may result in content removal, feature suspension, or account termination in accordance with our Acceptable Use Policy.

2. harm-prevention approach

We take a layered approach to preventing harm from AI-generated content, combining technical controls, system-level instructions, and human review:

System prompt guardrails. Every AI call includes instructions that frame outputs as symbolic, reflective, and entertainment-oriented — not authoritative guidance. Models are instructed to redirect sensitive topics (mental health crises, medical emergencies) to qualified professionals. Input validation. User prompts are passed through server-side validation before being submitted to AI providers. Requests that trigger safety filters are rejected before generation begins. Output content moderation. Responses are reviewed by OpenAI's own safety systems, which flag and refuse to produce harmful content categories including violence, CSAM, and self-harm instructions. User reporting. Every piece of content — including AI-generated readings shared publicly — can be reported by any user. Reports are reviewed by Tarosyn moderators within 24 hours. Iterative review. We monitor AI outputs for quality and safety issues and update system instructions in response to emerging patterns or misuse reports.

On a recurring schedule we run a structured fairness and safety audit of every AI surface, conduct a safety review before launching new AI features or materially changing prompts, and publish what we find in our AI Bias Report.

3. model output safeguards

The following technical and policy controls govern what our AI models are permitted to produce:

No medical, legal, or financial advice. System prompts explicitly prohibit the models from offering professional advice in these domains. If a user's input suggests they need professional help, the model is instructed to acknowledge the concern and recommend appropriate resources. Content limits for mature themes. Tarosyn is rated 17+. Mystic personas and reading interpretations operate within content parameters appropriate for that rating. Prompts that push toward explicitly sexual or gratuitously violent content are blocked. No factual claims about specific people. AI readings are grounded in symbolic and archetypal interpretation. Models are instructed not to make factual assertions about third parties (e.g., "your partner is cheating on you"). Hallucination framing. Outputs are framed as reflective insight rather than objective fact. Reading interfaces include notices that AI content is for entertainment and personal reflection. Model selection by risk level. We route requests to appropriately capable models: higher-stakes, user-facing content uses more capable models with stronger instruction-following; background and lower-risk tasks use lighter models. Template fallbacks. When AI generation fails or is unavailable, the platform falls back to curated template content rather than silently returning empty or malformed responses.

4. age-appropriate handling

Tarosyn is designed for users 13 years of age and older. Our age-appropriate AI practices include:

13+ age gate. All users must confirm they are 13 or older during registration. Users who cannot confirm an eligible age are removed from the platform. See our Privacy Policy §10 for details. Content rating compliance. AI outputs are constrained to content within our App Store and Google Play rating categories: Frequent/Intense Mature/Suggestive Themes (tarot, astrology), Infrequent/Mild Profanity (user-generated content), and Infrequent/Mild Horror. Outputs exceeding these bounds are blocked. No targeting of minors. We do not intentionally collect data about or direct AI personalization toward users under the minimum age. If we discover a minor has bypassed the age gate, we delete the account. Child profile features. The platform supports linked child profiles for family use. AI features available through child profiles are scoped to general astrological and zodiac content and exclude mature reading formats.

5. ai provider accountability

Tarosyn currently uses OpenAI as its primary AI provider. Our relationship with AI providers is governed by contractual and technical controls:

Data processor agreements. AI providers act as data processors under our privacy terms. They are contractually prohibited from using submitted content to train general-purpose models under enterprise/API usage terms. Minimal data submission. We send only the context necessary for a personalized reading: zodiac/chart positions, the selected mystic persona, the user's prompt, and the cards drawn. We do not send passwords, payment data, unrelated messages, or full contact lists to AI providers. Provider safety standards. We rely on providers who publish and enforce their own responsible-use policies, content policies, and safety red lines. We monitor provider policy changes and update our integration accordingly. Model version management. We pin or qualify the model versions we use and evaluate safety and quality implications before adopting new versions in production.

6. disclaimers & epistemic honesty

Tarosyn's AI features are tools for reflection, not authoritative sources of truth. We are committed to being transparent about what AI can and cannot do:

AI-generated readings are symbolic and interpretive. They do not predict the future with certainty and should not be treated as prophecy. AI content is generated by statistical models that can be wrong, biased, or inconsistent. Readings reflect patterns in training data, not objective facts about your life. Tarosyn's AI is not a substitute for professional medical, psychological, legal, or financial advice. In-app disclaimers appear on reading surfaces and in our Terms of Service. The "Truth State" dial (Veiled / Aligned / Unveiled) adjusts tone and candor — it does not change the factual accuracy or reliability of AI outputs.

7. reporting ai harms

If you encounter AI-generated content that you believe is harmful, inaccurate in a dangerous way, or violates this policy, you can:

Use the Report button that appears on every reading, Chant post, or shared content card to flag it for moderator review. Contact us directly at [email protected] with a description of the content and the concern. Review our Security Policy for responsible disclosure of any vulnerabilities that could be used to misuse AI features.

We review all AI harm reports and use them to improve our system prompts, filters, and provider configuration.

RELATED POLICIES

AI Transparency

The models we use, what data we send, and how readings are generated.

Responsible AI Principles

Our fairness, safety, transparency, and accountability commitments.

AI Bias Report

How we audit AI output for bias, what we found, and what we changed.

How your personal and spiritual data is handled when we process AI requests.

Acceptable Use Policy

Conduct rules that apply to all users, including AI feature misuse.

Community Guidelines

Standards for AI-generated content shared in the Chant feed and circles.

Questions? [email protected]