Phosphorus31 is a registered 501(c)(3) non-profit organization. EIN: 42-1888158
Responsible AI Human-Centered

AI Safety & Responsibility

Building artificial intelligence that respects human agency, protects privacy, and augments rather than replaces human capability.

Our AI Safety Commitments

P31 Labs has established clear principles for all AI development. These are non-negotiable requirements, not aspirational goals.

Privacy by Design

AI processing happens on-device whenever possible. No training on user data without explicit consent. No cloud retention of sensitive interactions.

Human Agency

AI suggests, humans decide. No automated actions without explicit approval. Easy override of any AI recommendation. Users remain in control.

Transparency

Clear disclosure when users are interacting with AI. Explanation of how recommendations are generated. No hidden persuasion techniques.

Accessibility

AI designed for neurodivergent users from day one. No manipulation of vulnerable populations. Support for executive function, not exploitation.

Technical Safeguards

Beyond principles, we implement concrete technical measures to ensure AI safety in our products.

1

Local-First Processing

PHOS and Centaur process everything possible on-device using WebLLM and similar technologies. No data leaves the device unless explicitly shared. Cloud AI is only used for complex queries with user approval.

2

No Training on User Data

We do not use user interactions to train or improve AI models. Your conversations with PHOS remain private and ephemeral. No data retention for model improvement.

3

Transparent Guardrails

Clear documentation of AI capabilities and limitations. Users understand what the AI can and cannot do. No anthropomorphizing or capability exaggeration.

4

Easy Opt-Out

AI features can be completely disabled. No reduced functionality for opting out. Core services work without any AI components. User choice is respected.

Centaur: Our AI Assistant

Centaur is our AI thought partner, designed specifically for neurodivergent users. It demonstrates our AI safety principles in practice—not just theory.

  • Suggests but never decides for you
  • Processes sensitive data locally
  • Clear about being an AI
  • Helps with executive function, doesn't exploit it
🐎
Human-AI Partnership

Ongoing Safety Research

We actively research AI safety issues relevant to our mission, contributing to the broader community while keeping our products safe.

Cognitive Support AI

Studying how AI can assist neurodivergent users without creating dependency or exploiting cognitive vulnerabilities.

Local LLM Security

Ensuring browser-based AI models don't leak data or introduce security vulnerabilities to client devices.

Voice Interface Safety

Preventing voice-activated AI from being triggered accidentally or manipulated through audio injection attacks.

Our Red Lines

We will never implement these AI capabilities, regardless of business pressure or user requests.

Emotional manipulation
Covert surveillance
Addictive engagement patterns
Autonomous decision-making

AI Governance

P31 AI Ethics Committee

All AI features undergo review by our internal ethics committee before deployment. This includes technical safety review, accessibility audit, and privacy assessment.

Mandatory safety review for all AI features
Quarterly AI safety audits
Public documentation of capabilities
User feedback integration

AI That Serves Humans

Experience AI designed with safety at its core. Centaur and PHOS demonstrate that helpful AI doesn't require surveillance or manipulation.