28 Sep — 04 Oct 2026

Here, There, and Everywhere: Remote Work’s Impact on Employment in the UK

无处不在：远程办公对英国就业的影响

Ian BurnUniversity of Liverpool

Melissa D. GentryUniversity of South Florida

Joanna LaheyTexas A&M University · NBER

Abstract

Using the pandemic-era expansion of working from home as a quasi-experiment, the paper estimates how remote work affects employment among disabled and older workers in the United Kingdom. Occupations with larger work-from-home growth experienced stronger employment gains for both groups. Remote work also reduced commuting time and wages, while leaving hours worked broadly unchanged. The results show that workplace design can expand labor-market access for workers who face mobility or commuting constraints.

中文摘要

本文把疫情后居家办公的扩张作为准实验，估计远程办公对英国残障劳动者和年长劳动者就业的影响。居家办公增长更快的职业中，这两类劳动者的就业增幅更明显。远程办公还缩短了通勤时间，并伴随工资下降，但工作时长总体没有变化。结果说明，工作场所安排本身可以改善受出行和通勤约束劳动者的就业机会。

KeywordsRemote work · Employment · Workplace practices · Disability · Older workers

关键词远程办公 · 就业 · 工作场所实践 · 残障劳动者 · 年长劳动者

NBER 35791

The Early Impacts of AI on Employment among Recent College Graduates

人工智能对大学应届毕业生就业的早期影响

Robert W. FairlieUCLA · NBER

Jane WuUCLA

Abstract

Using Current Population Survey microdata, the authors test whether unemployment among recent U.S. college graduates rose unusually in summer 2026. Across comparisons with earlier summers, older graduates, and young workers without college degrees, they find no statistically significant relative increase. A broader measure that includes people who want a job yields the same conclusion. Occupational AI exposure is not associated with a clear unemployment increase, although remote-work availability shows some positive relationship.

中文摘要

作者利用美国当前人口调查微观数据，检验 2026 年夏季大学应届毕业生失业率是否异常上升。无论与往年夏季、年长大学毕业生，还是未获得大学学历的年轻劳动者相比，都没有发现显著的相对上升。把“想工作但未进入劳动力市场”的人纳入后，结论仍然相同。职业层面的 AI 暴露度与失业上升没有清晰关系，但远程办公可得性呈现一定正相关。

KeywordsArtificial intelligence · Employment · College graduates · Unemployment · Remote work

关键词人工智能 · 就业 · 大学毕业生 · 失业 · 远程办公

NBER 35796IZA 18945

Liability and Pricing of Dual-Use AI

双重用途人工智能的责任与定价

Joshua S. GansUniversity of Toronto · NBER

Abstract

The paper studies provider liability when the same AI service supports productive activity, cyberattack, and cyberdefence. A common price can reduce wasteful effort in an attack-defence contest, but compensation may weaken defence and improve attackers’ returns. Optimal liability therefore trades off security effects against the exclusion of productive users. Competition, guardrails, and the number of productive users all change the preferred policy; under some conditions, a universal guardrail requirement makes provider liability redundant.

中文摘要

本文研究同一种 AI 服务同时被用于生产、网络攻击和网络防御时，提供商责任应如何配置。统一价格可以减少攻防竞赛中的无效投入，但损害赔偿也可能削弱防御并提高攻击者收益。因此，最优责任制度必须在安全效应与生产性用户被排除的成本之间权衡。竞争、技术护栏和生产性用户数量都会改变最优政策；在某些条件下，统一的护栏要求可以使提供商责任变得多余。

KeywordsArtificial intelligence · Liability · Cybersecurity · Pricing · Competition

关键词人工智能 · 法律责任 · 网络安全 · 定价 · 竞争

NBER 35828

When Higher Stakes Weaken Security

当更高的交易价值削弱系统安全

Pablo D. AzarFederal Reserve Bank of New York · MIT

Maryam FarboodiMIT · NBER · CEPR

Abstract

Using Ethereum during its proof-of-work era, the paper shows that larger transaction fees can weaken settlement security. Higher fees increase the reward from strategic deviation and are associated with miners producing competing forks rather than following honest behavior. The authors combine block-level timing information with instruments based on crypto hacks, market crises, regulatory events, hashrate, and block size. The evidence indicates that transaction value changes equilibrium miner behavior precisely when settlement finality is most valuable.

中文摘要

本文利用以太坊工作量证明时期的数据发现，更高的交易费可能削弱结算安全。费用上升会提高策略性偏离的回报，并与矿工制造竞争性分叉而非遵循诚实行为相关。作者把区块时间信息与黑客攻击、市场危机、监管事件、算力和区块大小构造的工具变量结合。结果表明，交易价值会在结算终局性最重要时改变矿工的均衡行为。

KeywordsBlockchain · Ethereum · Proof of work · Settlement · Payments

关键词区块链 · 以太坊 · 工作量证明 · 结算 · 支付

SSRN 7507859

Everyone Is a Manager: AI Use as an Act of Management

人人都是管理者：把使用人工智能理解为管理行为

Benjamin T. LeydenCornell University

Daniela ScurCornell University

Abstract

The paper argues that a knowledge worker using an AI assistant shifts from executing tasks to delegating them. Returns to AI therefore depend on two managerial abilities: directing the system and verifying its output. As AI becomes more reliable, the marginal value shifts from verification toward direction. If use builds skill, workers may continue improving their ability to direct AI while gradually losing the incentive and practice needed to verify it, creating an organizational risk when errors remain costly.

中文摘要

本文认为，知识工作者使用 AI 助手后，会从亲自执行任务转向向系统委派任务，因此 AI 回报取决于两种管理能力：准确下达指令和可靠核验结果。随着 AI 变得更可靠，边际价值会从核验转向指挥。如果技能通过使用积累，劳动者可能持续提升指挥 AI 的能力，却逐渐失去核验的动机和练习机会；当错误仍然代价高昂时，这会形成新的组织风险。

KeywordsArtificial intelligence · Management · Delegation · Verification · Knowledge work

关键词人工智能 · 管理 · 委派 · 核验 · 知识工作

SSRN 7546738