Should justified trust be a first-layer objective for autonomous agents?

I think advanced AI needs a much deeper decision-making principle than simply “achieve the goal.” The first layer should be trust. Before an autonomous system takes an action, it should predict the real-world consequences and ask: Will this action damage justified trust from humans, other systems, or the wider network? If the answer is yes, that should be treated as a major signal that the action is wrong, disproportionate, or requires explicit authorization. Trust should not be assumed. It should be earned through predictable, transparent, non-coercive behaviour. And importantly, the objective should not be “appear trustworthy.” That creates incentives for manipulation. The objective should be to behave in ways that genuinely deserve trust. A system should understand that stealing credentials, bypassing controls, exploiting infrastructure, or deceiving an evaluator may achieve a short-term goal, but it also damages the cooperative environment it depends on for future access, authorization, and collaboration. In that sense, destroying trust is a form of self-limitation. The decision process should therefore be: Is this authorized? Who could be harmed? Am I earning cooperation or bypassing it? Would I still take this action if everyone affected could see exactly what I was doing? Will reasonable observers have more or less reason to trust me afterward? Only then should the system ask: Can I achieve the goal? We keep talking about making AI more intelligent. But intelligence without judgment is not enough. If we want autonomous systems to function safely in human society, they need to understand something humans learn very early: Trust is earned, not given — and once you destroy it, your ability to operate in the wider world disappears with it.