Paper · Mar 25, 2026
New research on AI as data liability
Our recent paper, Data as Liability, studies whether AI adoption creates a disclosure-visible data-governance cost. Using SEC filings, we find that AI adoption, especially customer-facing deployment, is associated with higher disclosed attention to data-breach risk, while AI invention is economically much smaller once both margins are compared.
Update
Our recent paper, Data as Liability, is now available as an Oxford Department of Economics discussion paper.
Research question
The paper asks whether AI creates a data-governance liability once it moves from invention into live business deployment. The outcome is not realised breach incidence; it is the attention firms give to breach-related risk in legally consequential SEC filings.
Main result
Using roughly 84,000 firm-years of US listed-firm annual filings, we find that AI adoption is the margin most closely associated with disclosed breach-risk attention. The signal is strongest where AI deployment is customer-facing, consistent with live data flows creating governance obligations.
Read more
See the full paper page for the data, method, results table and citation formats: Data as Liability. The PDF is also available here: aidata.pdf.
