ING: Why “Autonomous AI” Is Misunderstood

In this conversation, Marnix van Stiphout of ING challenges one of the most loaded terms in today’s AI debate: autonomous. While the word is often used to signal risk, loss of control, or machines acting independently of humans, van Stiphout argues that this framing is misleading — and in many cases, unhelpful.

He begins by acknowledging the concern. Autonomous AI can sound alarming, particularly in a regulated industry like banking where trust, accountability, and control are paramount. But he is quick to reframe the discussion. Much of today’s IT landscape, he points out, is already autonomous in practice. Banks have long relied on systems that execute tasks automatically based on predefined rules written by engineers. From transaction processing to decision routing, applications “do things themselves” every day — not because they are uncontrolled, but because they are explicitly instructed to do so.

AI-driven agents, in his view, are an evolution of this same principle rather than a radical departure from it. The difference lies in speed, scale, and accessibility. AI allows these rule-based behaviours to operate faster, handle more complexity, and deliver benefits more directly to customers. But the underlying logic remains familiar.

Van Stiphout offers a practical example from lending and mortgage processes. Human employees operate within detailed work instructions that define exactly what they can and cannot do. Those same instructions can be passed to a digital agent. The agent becomes, in effect, a digital colleague — following the same boundaries, applying the same policies, and stopping when it reaches the limits of its authority.

Crucially, these systems are not unleashed to “run free.” Decision boundaries are clearly defined. When an agent reaches the edge of what it is allowed to decide, it must escalate to a human for intervention. Judgment, accountability, and responsibility remain firmly with people. In this sense, the agent is not autonomous in the way the term is often used — it is delegated, supervised, and constrained.

Van Stiphout’s message is one of precision and realism. The real risk, he suggests, is not AI acting independently, but misunderstanding how it actually works. By using language carefully and designing systems with clear guardrails, banks can harness the productivity and customer benefits of AI without surrendering control.

The future, he implies, is not about autonomous machines replacing humans, but about humans extending their capabilities through well-governed digital agents — tools that act quickly, consistently, and safely within boundaries we define.