衡量美国金融业的人工智能事件和影响（英）

A PROPOSED FRAMEWORK ABOUT OXFORD ECONOMICS Oxford Economics specializes in evidence-based thought leadership,forecasting, and economic impact analysis. Oureconomists use sophisticated analytical models and have access to a rich database of figures, forecasts, and analysison 200 countries, 100 industrial sectors, and 8,000 cities and regions.Headquartered in Oxford, with offices aroundthe world, we employ more than 600 people, including over 400 economists, industry experts, and business editors. ABOUT HCLTECH HCLTechis a global technology company, home to more than 226,600 people across 60 countries, deliveringindustry-leading capabilities centered around AI, digital, engineering, cloud and software, powered by a broadportfolio of technology services and products.We work with clients across all major verticals, providing industrysolutions for Financial Services, Manufacturing, Life Sciences and Healthcare, High Tech, Semiconductor, Telecom CONTENTS INTRODUCTION..........................................................................................................................................3SECTION 1.AI RISK IS A GROWING CONCERN, BUT REPORTING IS INSUFFICIENT............4SECTION 2.DEFINING A FINANCE-SPECIFIC TAXONOMY OF AI INCIDENTS........................6 INTRODUCTION Artificial intelligence (AI) has been part of the financial industry technology stack fordecades in the form of machine learning, predictive analytics, and rules-based automation But the emergence and rapid deployment of generative and agentic AI has expanded the scope,autonomy, accessibility, and unpredictability of AI systems. This shift significantly broadens therisk landscape for companies using these advanced tools. A standardized framework that Quantifying the impact of AI incidents requires a methodology for categorizing their type andmeasuring their frequency, severity, and costs—both direct and indirect, and financial and non-financial.Few such frameworks exist today,but documentation efforts from established, Commercial and academic research into organization-or sector-wide impacts is also sparse atthis stage. Still, early studies on this topic make clear the business risks posed by AI, importance OxfordEconomics,in collaboration with HCLTech,reviewed existing literature and recentSecurities and Exchange Commission (SEC) filings to understand how AI-related incidents areemerging, classified, and disclosed within the US financial services sector. Our research points togrowing concernsabout AI risk and a lack of transparency around incident reporting.A SECTION 1. AI RISK IS A GROWING CONCERN,BUT REPORTING IS INSUFFICIENT cybersecuritycategories.3Further, public companiesareonlyrequired todiscloseincidentswhen theydeterminethe issue is material under SEC standards; The SEC has issued companies repeatedwarnings to accurately disclose AI-relatedrisks and taken severalenforcement actionsagainst those engaged in “AI washing,” orthe exaggeration and misrepresentation of Analysis conducted by Business Insider with the AI-poweredmarket research platform AlphaSensefound that 418firmsworthmore than $1 billionhave cited AI-relatedreputational risksin their SECfilings—up46% from 2024 and about nine times Alarge-scale,systematic analysis of AI riskdisclosures in SEC 10-K filings conducted by the LawandTech Lab at Maastricht University in theNetherlands—a preprint study, but the first of itskind—found that mentions of AI risk increased from4% in 2020 to 43% in 2024.2Yet the study notes Based on our search, financial institutions do notexplicitly describe specific AI-related incidents theyhave encountered or successfully mitigated. Rather,their brief statements focus on AI-related products;AI-related risks from third parties, changing lawsandregulations,operational errors,and cyber increase in$1bn+companies citingAI-related reputational risks in SECfilingssince2024Business Insider / AlphaSense 1U.S. Securities and Exchange Commission,SEC ChargesTwo Investment Advisers with Making False and MisleadingStatements about Their Use of Artificial Intelligence, March18, 20242Lucas G.Marin, Bram Rijsbosch, Gerasimos Spanakis, andKonrad Kollnig,Are Companies Taking AI Risks Seriously? A benchmark, or aggregate incidents within and Increaseddisclosures of risk do not necessarilyindicate a proportional rise in actual incidents, butthelack of transparency creates substantial •Minimal or inconsistent information on incidentseverity, impact, financial cost, and metadata, aswell as mitigation or remediation efforts, which •Overemphasis on social impacts over business •Reliance on crowdsourcing and web scraping,which creates information asymmetry due toinconsistent or absent requirements to publiclydiscloseAI incidents;this leads to over- InonepreprintstudyexaminingAI-risktransparencyin financial markets,the authorsreferencean International Money Fund(IMF)observation that the August 5, 2024, equity selloffprovides an instructive ex