Trustworthy AI

The deeper I research, the more I believe that using AI well and using AI safely have to travel together, especially as the rules tighten.

What trustworthy AI is

Picture certifying a car for the road. Going fast is not enough. It needs brakes, seatbelts, lights, and it has to pass inspection. Trustworthy AI is the same idea: a usable AI system needs more than intelligence. It has to be transparent, fair, accountable, protective of data, and controllable.

In 2019, the European Union's high-level expert group on AI published seven requirements for trustworthy AI, which became the foundation for the governance framework that followed. In August 2024 the European AI Act came into force, sorting AI into four risk levels: prohibited, high risk, limited risk and minimal risk, with obligations tightening as the level rises.

How this applies to running a business

I find many Vietnamese businesses think these rules are far away. But if you sell, partner or collect data touching international markets, that rulebook reaches you sooner than you would expect. And even with no legal obligation, an AI system that is transparent and fair is what keeps customers' trust.

What is worth doing right away is sorting: which tasks you hand to AI are low risk, such as suggesting content or summarising, and which are high risk, such as anything affecting entitlements, money, or decisions about people. The higher the risk, the more transparency and clear accountability it needs.

If you sorted the work you currently give AI by risk level, which is the highest-risk one you have not put enough brakes on?