2025年Elliptic报告：预防加密资产生态中 AI 驱动犯罪的最佳实践

Executive Summary Throughout 2024, Elliptic conducted horizon scanning work to understand and devise best practices toprevent AI-enabled crime trends in the cryptoasset ecosystem. The aim is totackle emerging risksearly on and prevent them from becoming more mainstream, while understanding how theblockchain analytics sector can support stakeholders to mitigate threats. This report details the results of this consultation, which involved 40 participants from a range ofindustries, including law enforcement, virtual asset services, regulators, tech startups and academia. Key findings include: AI-enabled crime trends, including deepfake scams and AI-enabled illicit goods and services, areprojected to become more mainstream in the next three years, with notable impact. •Various stakeholders will need to play to their strengths and adopt a series of complementaryprevention measures to ensure that the risk does not overwhelm resources, hinder legitimateconsumers or slow downthebeneficial innovation of AI/crypto technologies•This will involve upscalingcapacity to detect and mitigate AI-enabled crime risks by employingdefensive AI capabilities and enforcing clearer expectations on enablers such as social media•Better authentication systems to safeguard against malicious uses of AI chatbots, socialmedia accounts and crypto services will also be necessary to circumvent deepfakes, illicitprompts and other criminal innovations•Enhancedcooperation between and across stakeholders, including cross-border data andknowledge sharing, will be crucial to tackle the industrialized transnational nature of some AI-enabled risks, in particular crypto investment/romance (so-called “pig butchering”)scams•Regulators will be at the forefront of working with, not against,industry and ensuring theimplementation of best practices in a balanced and feasible manner This report details how consultedparticipants viewed the current prevalence of AI-enabled cryptocrime risks and the future likelihood and impact of thembecoming mainstream. It also analyses their suggestions for best practices and associated ratings for their perceivedeffectiveness and monetary/social costs of implementation. This report is intended to be a practical guide for stakeholders–ranging from law enforcement tocompliance professionals–seeking to protect their institutions from AI-enabled crypto crime threats. Introduction In June 2024, Elliptic released its AI-enabled crimes in the cryptoasset ecosystem report. This was thefirst part of a horizon scanning exercise designed to: 1.Understand current and emerging crypto crime risks exacerbated by AI2.Identify prevention measures, vetted by various industries, to ensure the safe and sustainableinnovation of both AI and blockchain technologies–unimpeded by crime For the second aim, we launched a cross-industry consultation that asked various experts andstakeholders about their current experience with AI-enabledcrypto crime and their preferred meansof preventing them. This briefing note sets out the results of this exercise. The consultation was conducted in the form of a Delphi study–a futures-oriented surveymethoddeveloped by the RAND Corporation in the 1950s to gauge expert views while building consensus. Youcan find out more about this methodology in the Appendix. Encouraged by theincreasinglypro-crypto approach ofnumerousjurisdictions, our aim through thisresearch is to engage various stakeholders while the threats remain relatively in their infancy. Thisallows forearly actionto prevent them from becoming mainstream, thereby remaining ahead of thecurve in our fight against tech-savvy criminals. To reiterate, Elliptic’s approach to emerging crime trends is guided by the UK Government Office forScienceFutures Toolkitand the4P approach, namely “pre-empt, protect, provide, promote”–visualised below. This reportfirstintroduces the participants consulted and the results of the cross-industry consultation–specifically how they rated the risk of AI-enabled crypto crime trends and their proposed preventionmeasures. Best practices are introduced based on howwellthese measures werereceived. Participants This study, per the Delphi methodology, took place over three rounds, as shown in Table 1. The background of participants showed ahealthy distribution across law enforcement, virtual assetcompliance and research institutions (e.g. think tanks and universities), as well as regulatory agencies,AI startupsand FinTech. Participants hailed from every inhabited continent in the world. The figure below shows thebackground and jurisdiction of the participants. Emerging risks and trends This study, per our June 2024 report, fielded 16 emerging AI-enabled crypto crime trends–categorizedinto five typologies–forparticipants to rate from 1 (low) to 7 (high) according to: 1.Prevalence:the extent to which thecrimetrend is currently being observed2.Likelihoodof thecrimetrend becoming mainstream in thenext three years3.Impact: