Artificial intelligence adoption and the term structure of corporate default risk

Purpose This paper provides evidence on whether corporate investment in artificial intelligence (AI) predicts corporate default risk, using the full term structure of forward-looking probabilities of default (PD). Design/methodology/approach Validated firm-level AI-investment measures built from employee resumes and job postings are merged with NUS-CRI forward-intensity PDs for U.S. listed non-financial firms (outcomes 2011–2018). Because PD is a fraction in [0,1], estimation uses fractional logit with industry and year fixed effects; identification is probed with a university AI-talent-supply instrument in a control-function design with bootstrap inference for the generated regressor, matching and entropy balancing and an extensive robustness battery. Findings A one-standard-deviation increase in AI investment is associated with significantly lower default probability at 12- to 36-month horizons – about 8 basis points at 12 months, roughly one-quarter of the sample mean. The relative effect declines with horizon and is concentrated within the first two years. A control-function test does not reject exogeneity, so the fractional-logit estimates are preferred, with the instrument as corroboration; results are robust to the job-posting measure, alternative industry classifications and screens and R&D and capital-intensity controls. Originality/value To our knowledge, the first firm-level evidence linking validated AI-investment measures to the PD term structure for U.S. listed firms, showing that labor-market footprints of AI adoption carry default-relevant information and that the association is a short-to-medium-horizon phenomenon.