Regulatory sandbox

The idea comes from financial regulation and has spread to AI. Instead of banning a system or approving it blind, the regulator lets a company run it for a set period, on a limited scale, while reporting what happens.

Poland’s digital affairs minister announced sandboxes alongside a new AI commission after the Warsaw robot demonstration. In practice that could mean a firm testing an AI hiring tool on a few hundred applicants with the regulator watching the outcomes, rather than rolling it out nationwide first.

Critics say sandboxes let companies buy time and legitimacy. Supporters say they produce evidence regulators otherwise never see.