The public regulation of the chemical industry rests on the use of IT tools that were developed to predict the most common toxicities associated with various families of molecules. Public regulatory authorities and chemical firms have a joint interest in using such predictive tools, namely the fact that these tools reduce the amount of animal experimentation to perform before allowing chemicals on the market, making it less time consuming, costly, but also more ethical. Of all of the tools used, some are more strictly speaking “predictive”: they allow for simulating highly specific, quantitative models, to numerically determine the toxicity of new molecules. Publicly-developed tools are of a different sort: they are more of a screening platform, allowing to compare data concerning series of molecules, to produce a hypothesis about the likelihood that a newly developed molecule, comparable to past ones, develop one sort of toxicity or another. In other words, commercial tools seem to respond to the wishes of the chemical industry to reach a high level of precision, to adjust (and minimize) regulatory intervention on their products. Publicly-developed tools aim less for precision, but more for reliable and credible simulations, allowing to take firm, non-contestable decisions. Drawing on these observations about differing epistemologies underlying public or private tools, this paper will explore what one may call the political economy of prediction, verifying if the broader political economy of chemicals regulation, said to be one of industry dominance, reflects in dominant ways of predicting.