arXiv Science⌕ Search

arXiv · 2610.14305

Safety Research and the Allocation of Frontier Compute

Abstract

During a frontier takeoff period, an AI lab splits a committed stock of compute among capability research, safety research, and customers, and a regulator limits both the services customers lose and how long any capability can be delayed. Extra safety compute can then come from customers or from capability research. Funding from capability research delays the end of the takeoff period, so the cheaper source depends on how much finishing raises customer services: customers above a threshold, capability research as well below it. Customer funding leaves capability dates unchanged, while research funding gives safety knowledge more time at the cost of later capability and customer services. The threshold is evaluated at a given allocation under a common continuation. Varying the calibration shows how delay costs, additional safety learning, and service valuation move it; in the baseline, it rises with the service budget. Customer-funded safety is front-loaded relative to customer use under the baseline timing conditions. Within the tested constant-share rules, commitment before the safety technology is known costs little unless safety work also advances capability. At the baseline budget, front-loaded customer-funded safety cuts catastrophe probability from 75 to 51 percent when full-speed catastrophe has a 75 percent probability, and from 25 to 14 percent when it has 25 percent. Either way it comes within one point of the best tested rule that also cuts capability research, and the funding threshold does not move.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Li Gan. 2026-11-05. Safety Research and the Allocation of Frontier Compute. https://arxiv.science/abs/2610.14305

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Just After Minimum Wage Hikes: Short-Run Labor-Demand Response and Reallocation

How labor markets adjust immediately after minimum wage hikes remains an open, policy-relevant question. This paper studies short-run minimum-wage effects in Japan's spot labor market using Timee data and a wage-bin difference-in-differences design. We find a normalized net employment-share change of $-0.010$, with similar estimates for vacancies and total working hours. Weekly and daily estimates show that these effects emerge immediately after implementation and then stabilize. Effects are more negative where the minimum-wage bite is higher and in restaurant work. Net share effects are negative for midnight shifts and postings offering transportation reimbursement. Gender-targeted and beginner-targeted postings show no significant net change. Decomposing vacancies by employers' posting histories, we find that incumbent employers cut postings below the new floor while employers that did not post before the revision post just above it. Cross-employer reallocation, a multi-year process in standard labor markets, thus begins immediately in spot gig work.

econ.GN↗

AI Adoption in S&P 500 Firms

The adoption of artificial intelligence (AI) by large enterprises is an important potential source of aggregate productivity improvement and labor market impact. We study AI adoption of S&P 500 firms over the period 2016 to 2025, estimating adoption at the enterprise level. While generative AI tools are useful for personal and professional applications, our focus is on the deep integration of AI in the business processes of large enterprises which are bellwethers for firm adoption more broadly. We develop a novel measure to assess deep AI adoption (and distinguish it from AI hype) that is based on SEC 10-K filings, where laws and regulations ``prohibit companies from making materially false or misleading statements." In 2025, 11% of S&P 500 enterprises had AI deeply integrated into their business processes, and a further 10% were using AI in the production of goods and delivery of services. AI adoption has more than quadrupled from 5% in 2022 with slowly accelerating adoption among non-technology firms but very aggressive adoption in the technology sector which accounts for two-thirds of deeply integrated enterprise adoption. Firm profitability shows a "J-curve" as firms move from no adoption to deep adoption, but we observe no differences in capex or productivity. Among technology firms, but not others, AI adoption is higher for firms with more employees and higher values of Tobin's q.

econ.GN↗

Clicks, Carbon, and Risk: Climate Attention, Persistent Financial Risk, and Real-Sector Activity

Do shifts in public climate concern alter persistent financial risk and subsequently matter for real activity? We combine daily U.S. sectoral equity returns with a monthly Climate Concern Index (CCI) constructed from internet search intensity and decomposed into adverse and transition-oriented components. GARCH-MIDAS estimates reveal substantial heterogeneity: adverse climate concern raises persistent volatility in carbon-intensive Energy activities, while transition-oriented attention tends to reduce persistent volatility in regulated Utilities, with important differences across sub-sectors. We then construct a persistent-risk differential between Energy and Utilities and examine its predictive content for sectoral real activity using local projections. Innovations associated with adverse climate concern precede declines in Oil and Gas industrial production and employment over the following quarters, while comparable Utilities responses are limited. The production result is preserved when the financial predictor is reconstructed using the more closely matched Oil, Gas, Coal equity sub-sector and remains economically similar across several robustness checks. Overall, the findings establish a link between climate concern, persistent sectoral financial risk, and subsequent real-sector activity.

econ.GN↗