Responsible AI means more than building high-performing AI. It means designing, using, and governing AI in ways that are acceptable to people and society. Six widely recognized principles are Fairness, Transparency, Accountability, Safety, Privacy, and Human Oversight.

Fairness means ensuring that AI decisions do not impose unreasonable disadvantages on particular individuals or groups. Because AI may inherit existing biases from training data or system design, it is necessary to examine not only whether outcomes differ, but whether such differences have reasonable justification.

Transparency means enabling people to understand and verify how and under what conditions AI decisions are made. It does not require every internal process to be fully explainable, but users and supervisors should at least be able to trace the basis of a decision and the rules applied.

Accountability means ensuring that responsibility for AI decisions is not left ambiguous. Responsibility should not be attributed to AI itself; the roles and responsibilities of developers, providers, deploying organizations, and operators must be clearly defined.

Safety means preventing harmful outcomes arising from AI use. It includes not only safeguards at the design stage, but also mechanisms for detecting anomalies or unexpected effects during operation and stopping or correcting the system when necessary.

Privacy means appropriately protecting personal and sensitive data. Data must be managed according to defined purposes and permissions throughout its lifecycle, including collection, use, storage, and sharing.

Human Oversight means ensuring that important decisions are not left entirely to AI and that humans retain meaningful capacity to intervene, review, and correct them. Human involvement must be substantive rather than merely procedural, with access to the information and authority necessary to make actual decisions.

These six principles provide an important foundation for Responsible AI. Principles alone, however, cannot ensure that AI systems actually operate in accordance with them. Questions such as how Fairness is evaluated, who bears Accountability, and when Human Oversight is required must be translated into concrete governance mechanisms. The Governance OS addresses how these desirable principles can be connected to actual AI decision-making and execution.