Quantitative Finance · Thu, Nov 5, 2026

arXiv Quantitative Finance (q-fin).

RSS feed
arXiv cs.AI · cs.HC · econ.GN

Dull, Dirty, Dangerous? Working Conditions and the Limits of Occupational AI Exposure

In briefDull, Dirty, Dangerous? How Working Conditions Relate to AI Exposure

SummaryThis paper uses 21 O*NET items across 754 US occupations to examine how dull, dirty, and dangerous working conditions relate to occupational AI exposure. Higher exposure is associated with lower condition scores under both an LLM-based measure and AIOE. After adjusting for physical demands, education, training and experience, and wages, the composite association shrinks, with the LLM-based confidence interval including zero.

Siyi Yu

arXiv econ.GN · cs.AI · q-fin.EC

Does Frontier AI Strengthen Cyber Defense? Evidence from Project Glasswing

In briefDoes Frontier AI Make Cyber Defense Stronger? Evidence from Project Glasswing

SummaryThis paper studies the effects of giving a small group of large US tech firms early access to Claude Mythos Preview, an AI model for finding software vulnerabilities, under Project Glasswing. Firms with access more than doubled their publicly disclosed vulnerabilities and disclosed about 150 percent more high-impact flaws, while fix speed stayed the same. The share of vulnerabilities exploited within a month of disclosure fell by about two thirds, while the number exploited did not change.

Thomas Mosk