The 4 founding principles
ACF® rests on four founding principles, independent of the technical protocol used underneath. They apply to any agentic system in production, regardless of the client (Claude, GPT, Gemini, proprietary model) or the protocol (MCP, ACP, Operator, etc.).
P1 · Decision Sovereignty
Critical strategic decisions are never delegated to autonomous agents. The organisation keeps ultimate accountability for the decisions its agents take, whatever level of autonomy it delegates to them. An autonomous AI agent never moves decision accountability outside the organisation. Operational delegation (the agent acts) does not entail responsibility delegation (the organisation answers).
P2 · Governance by Design
Governance frameworks are defined before deployment, not after, and their intensity is proportionate to the criticality of the decisions entrusted to the agent. An agent is not put into service before its framework exists: agentic constitution (cf. ACF-04), mandate (cf. ACF-13), criticality matrix (cf. ACF-03) and stop mechanism (cf. ACF-07). This framework is proportionate: an agent that only suggests does not warrant the same control apparatus as an agent that decides on its own. Over-investing controls on a low-criticality decision wastes resources; under-investing on a critical decision creates regulatory and operational exposure.
P3 · Ultimate Human Control
Every agentic system must preserve the ability for human intervention: a takeover mechanism that can be activated at any time and is calibrated to criticality. An agent's autonomy is a conditional privilege, never a vested right. At any moment, an authorised operator must be able to freeze, redirect or revoke the agent. This is the operational expression of AI Act Article 14.
P4 · Traceable Accountability
Every autonomous action must be auditable and attributable: any decision can be reconstructed after the fact by pointing to the doctrine, rules and data that produced it, and to the person who authorised it. An agentic decision without a reconstructible audit trail is legally indefensible. ACF® mandates that, at any time T, the organisation can answer the question: 'which doctrine version, which rules and which data produced this decision?'
Why axiomatic?
ACF® principles are posited as axioms, not derived from anything else, so an organisation can recognise or reject them as a whole. A debate on “should we trace decisions?” does not lead to a deployment. A decision on “do we adopt the ACF® framework?” unblocks the next step. This is a deliberate methodological posture.