Safety is the foundation.
We develop AI systems that are transparent, controllable, and aligned with human values. Safety isn't a feature—it's the requirement for deployment.
The Safety Charter
This document serves as the constitutional basis for our AI development. It overrides commercial incentives and binds our leadership to specific safety outcomes.
VIEW_FULL_DOCUMENT_V1.PDFConstitution of
AI Safety
Our AI systems will never knowingly deceive users or misrepresent their capabilities. Transparency is the default state. Systems must identify themselves as artificial entities upon request.
Humans will always retain the ability to override, correct, or shut down our systems. We build tools that empower human agency, not replace it. "Kill switches" must be hardware-gapped.
0x7F4A...9C2BSystem Axioms
These principles act as the kernel-level logic for our organization. They guide every decision from research to deployment.
Human Agency
AI systems must remain subservient to human command. Override protocols are hard-coded into the inference engine.
Radical Transparency
We disclose capabilities, limitations, and failure modes. We do not anthropomorphize or deceive users.
Staged Deployment
Capabilities are released incrementally. We employ a "sandbox-first" approach to verify safety boundaries.
Harm Refusal
Models are trained via Constitutional AI to proactively refuse requests involving violence, hate, or illegal acts.
Adversarial Hardening
Continuous Red Teaming. We pay hackers to break our systems before they are deployed to the public.
Public Accountability
We submit to third-party audits and publish our safety methodologies for peer review.
Report a safety concern.
Identified a safety risk, jailbreak, or policy violation? We provide protected disclosure channels and respond to critical reports within 24 hours.