Risk Ceilings and Development Deadlines: Pacing AI under Uncertain Safety Productivity

Li Gan

Home/Archive/arXiv

arXiv · 2610.11093

Risk Ceilings and Development Deadlines: Pacing AI under Uncertain Safety Productivity

Li Gan

View original source ↗Read source document ↗Export metadata ↓

Abstract

Can a regulator promise both a risk ceiling and a development deadline when safety productivity is unknown? A ceiling below the final model's unprotected hazard requires a minimum stock of safety knowledge, so both promises hold only if weak research can be ruled out. Learning first and then replaying development, with spare compute in safety, comes close to that minimum. In a calibration with only state risk, constant safety yield, and full knowledge transfer, it needs only 3.3 percent more productivity than the necessary bound. Customer services pay for the guarantee. Rules that fix compute allocation fix dates, not risk.

Explore related subjects

econ.GN cs.CY q-fin.EC

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Li Gan. 2026-10-08. Risk Ceilings and Development Deadlines: Pacing AI under Uncertain Safety Productivity. https://arxiv.org/abs/2610.11093

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Delphos: A reinforcement learning framework for assisting discrete choice model specification

We introduce Delphos, a deep reinforcement learning framework for assisting discrete choice model specification process. Delphos aims to support the modeller by providing automated, data-driven suggestions for model specifications, thereby reducing the effort required to develop and refine utility functions. Delphos conceptualises model specification as a sequential decision-making problem, inspired by the way human choice modellers iteratively construct models through a series of reasoned specification decisions. In this setting, an agent learns to specify candidate model specifications by choosing a sequence of modelling actions, such as adding alternative specific constants, accommodating both generic and alternative-specific taste parameters, applying non-linear transformations to attributes, and including interactions with covariates. Each resulting candidate model is estimated and evaluated using a reward function defined by the modeller, which can reflect statistical model fit as well as behavioural expectations. Specifically, Delphos uses a Deep Q-Network to learn how individual specification decisions contribute to the eventual quality of the resulting model and, in turn, which sequences of modelling decisions tend to produce well-performing candidates. We evaluate Delphos on both simulated and empirical datasets using alternative reward functions. In simulated cases, learning curves, Q-value patterns, and performance metrics show that Delphos learns effective specification strategies while exploring only a small fraction of the feasible modelling space. We further apply the framework to two empirical datasets to benchmark and demonstrate its practical use. These experiments illustrate the ability of Delphos to generate competitive, behaviourally plausible models and highlight the potential of this adaptive, learning-based framework to assist the model specification process.

econ.GN↗

arXiv2026

Trade Liberalization and Product Innovation: The Dynamic Role of Exporting

How does trade liberalization affect firms' incentives to innovate? We answer this question using China's WTO accession and a dynamic model of firms' joint export and product-innovation decisions. We find that trade liberalization reduced iceberg trade costs by approximately 13.5 percent in the air-conditioner manufacturing industry, which substantially increased the probability of export and product innovation. Moreover, the response is driven primarily by dynamic rather than static incentives. Two dynamic mechanisms are central: exporting and innovating lead firms to predict more favorable transitions to better productivity states and reduce future entry costs, thereby raising the returns to subsequent innovation. Counterfactual decompositions show that these mechanisms account for the majority of the innovation response. For firms in an intermediate productivity state, the direct static effect (without entry-cost saving and state transition) accounts for only 9.2 percent of the total effect of WTO accession. Shutting down state transition leads to 62.4 percent of the total effect, while shutting down entry-cost saving results in 28.5 percent. The results demonstrate that the innovation effects of trade liberalization are substantially amplified by firms' endogenous dynamic responses, highlighting the importance of accounting for state transitions and forward-looking decisions when evaluating the gains from trade.

econ.GN↗

arXiv2026

Who Leads and Who Collects:Algorithmic Collusion in Markets of Heterogeneous Language Models

Evidence that pricing algorithms collude comes from markets in which every seller runs the same algorithm. We ask what happens when they do not. Four language models from four providers, each at its cheapest tier, price in a four-firm logit Bertrand market without communication, in every homogeneous, two-by-two and fully mixed composi- tion (11 cells, 20 runs, 200 periods). Collusion is a property of the model: Claude and Gemini markets reach 72 to 79 percent of the monopoly rent, DeepSeek markets 24 percent, and GPT markets none, though GPT prices drift above the monopoly level rather than toward competition. Mixing does not reduce collusion by itself. Markets containing Gemini, which opens at the highest price and settles highest, are more col- lusive than the homogeneous markets they are built from; markets containing Claude, which opens lower and follows its rivals down, are less so; and only the fully mixed market is significantly less collusive than the average homogeneous one. Stability de- pends on the least stable participant: two GPT firms suffice to keep any market from converging. Inside mixed markets the rent is shared in a transitive order, DeepSeek over Claude over Gemini over GPT, that inverts the anchor ranking. The model that raises the price collects the least of the rent, as the price-leadership model predicts.

econ.GN↗