The regulatory framework for emerging algorithmic systems often oscillates between the permissive 'innovation-first' model and the restrictive 'precautionary principle.' Under the precautionary principle, developers must provide rigorous evidence of a system's safety and non-bias before public deployment. Proponents argue that this ex-ante oversight is essential to prevent systemic harms that are difficult to rectify post-hoc. Conversely, critics contend that the burden of proof is so high that it stifles the development of beneficial technologies, particularly in fields like diagnostic medicine where the cost of delay is measured in human lives. A more nuanced approach, currently gaining traction in science policy circles, involves risk-stratification. Rather than applying a uniform precautionary standard, this model adjusts the evidentiary burden based on the system’s domain of application and the reversibility of its potential errors. For instance, an algorithm used for low-stakes inventory management would face minimal scrutiny, whereas one governing credit allocation or autonomous transit would require extensive validation. By decoupling the level of oversight from the technology itself and attaching it to the specific socioeconomic outcome, regulators can protect public interests without imposing a moratorium on technological progress.