From the 1970s onwards, a growing number of national and international organizations have been working to generate a definition of precisely what constitutes a carcinogenic substance. The central question for these new groups of experts then became what type of data the authorities should rely on to govern chemicals and their uses.

I focus on the discussions about whether or not animal bioassays are proof of carcinogenicity for humans. I explore those discussions and negotiations within the institutional network of the International Agency for Research on Cancer (IARC), an organization linked to the World Health Organization that has classified chemicals for their carcinogenicity since the 1970s.

Drawing on a fresh set of archives, I show that over the 1970s and 1980s, these criteria for evaluating chemical substances became a resource for scientific and regulatory institutions to stabilize the relations between themselves. I then show how those institutional relations have been challenged by the rise of a new kind of data, called "mechanistic data," which has been progressively integrated into the IARC evaluation criteria.

This presentation thus sheds light on the large transformations that have affected the definition of toxic substances during the recent historical period.

From the 1960s to the 1980s, US officials sought to reduce cancer incidence by regulating chemical exposures. However, rodent carcinogenicity tests were lengthy and expensive, making identifying the carcinogens among 60,000 chemicals on the market impractical. Introduced in 1973, the quick and inexpensive “Ames test” indicated whether a chemical caused mutations in bacteria, and putatively cancer in humans. In the 1970s and 1980s, scientific societies, international bodies, and government agencies launched efforts to validate the Ames test for carcinogenicity. While the Ames test became the most commonly used mutagenicity (or genotoxicity) test, technical determinations of its validity could not resolve policy issues concerning how much predictiveness was required for regulatory action. Furthermore, the availability of new animal carcinogenicity data through the 1980s in turn affected the validation of the Ames test. Examining the challenges with validating even a very successful chemical safety test reveals the limits of what one can expect from regulatory science.