DEME 2.0: Real-Time Ethical Governance for Safety-Critical Autonomous Systems via Computable Moral Landscapes

We present a computational architecture for in safety-critical autonomous systems that must make high-stakes decisions at while remaining transparent and auditable to humans. DEME~2.0 treats ethics as a , representing decision-making as navigation over a . Each candidate action is mapped to a moral vector whose coordinates encode ethically salient quantities such as expected harm, rights respect, fairness, autonomy and epistemic quality. ---specified by regulators, institutions or communities---interpret this landscape through hard veto regions, lexical priorities and scalarization functions that can be debated, versioned and combined across stakeholders. A central contribution is a that compiles governance profiles into hardware-resident Ethics Modules capable, in principle, of enforcing non-negotiable and ranking permissible actions within the of control loops (sub-millisecond and potentially sub-microsecond budgets) on contemporary embedded hardware. We show that profile validation, priority consistency checking and runtime decision resolution all admit , ensuring computational tractability even for rich governance structures. A generates tamper-evident decision proofs, linking high-level stakeholder values directly to machine-speed outcomes and aligning with traceability requirements in emerging regulatory regimes such as the EU AI Act and NIST AI RMF. Conceptually, DEME~2.0 can be viewed as a proposed in moral philosophy: moral peaks and valleys are instantiated as coordinates in a high-dimensional vector space, and governance profiles become algorithms for moving through that space. By bridging moral philosophy, formal methods and embedded systems engineering, DEME~2.0 provides a foundation for in contested moral terrain. Author preprint deposited for archival and citation. Draft — pending author review.