Security, Privacy and AI Safety
Download PDF- Data minimization, encryption, key separation, access control and purpose-based storage isolation are applied by default.
- High-risk permissions are subject to multi-party approval, time delays, limits, re-authentication and an Emergency Pause.
- Contracts, the Gateway, reserves and smart wallets undergo independent security audits and monitoring.
- Model outputs are continuously evaluated for accuracy, bias, over-reliance, prompt injection and data leakage.
- Mental care features include crisis guidance and referral procedures to professional organizations, and restrict definitive medical statements.
- Privacy impact assessments and AI risk assessments are built into product launch gates.
Privacy-enhancing technologies such as selective disclosure, cohort thresholds, zero-knowledge proofs, TEEs and confidential computing are being evaluated in stages. Because applying a technology does not by itself guarantee privacy, each is assessed together with the threat model, operating procedures, key management and legal accountability.