AI transparency is now operational. Agentic control must follow.

Another part of AI governance is now operational

On 2 August 2026, another part of AI governance moved from policy into operation. Transparency obligations under Article 50 of the EU AI Act now apply to relevant providers and deployers. The European Commission's enforcement powers for general-purpose AI obligations have also entered into application.

These are significant steps. People should know when they are interacting with AI and when certain content has been generated or manipulated by it.

Transparency is only one boundary

As AI moves from producing content to initiating payments, changing software, operating machines or influencing consequential decisions, organisations face a deeper question: what was the system authorised to do—and can that authority be demonstrated afterwards?

A probabilistic model can produce valuable judgements, but it does not become a dependable control system merely because its output is labelled. Agentic AI connects model outputs to tools and real-world actions. That connection requires explicit limits, escalation paths and durable evidence.

Intelligence and authority are different

An AI system may propose an action. A governed system must determine whether that action is permitted in its particular context. It must also preserve sufficient evidence to examine what happened, who requested it and which authority applied.

This is especially important in finance, robotics, drones, critical software and other environments where an action can create material consequences.

Governance should enable AI to go further

PF Systems’ position is that organisations should not have to choose between capable AI and accountable operation. Governance should enable AI to go further by controlling the boundary between a probabilistic recommendation and an authorised action.

Within PF OS, that responsibility is deliberately separated. PF Memory knows. PF Core proves. PF Kernel decides.

This is not a claim that governance makes the underlying AI deterministic, eliminates risk or automatically establishes compliance. It is an architectural response to the last mile of agentic computing: the point where intelligence becomes action.

The AI Act's latest milestone reinforces a wider commercial reality. AI transparency is becoming expected. For consequential agentic systems, demonstrable authority and evidence will increasingly matter too.