理解 AI 如何进入法律服务的交付、专业分工与组织运营,并据此观察不同实践。Understand how AI enters legal-service delivery, professional roles and organizational operations, and use that lens to examine different practices.
什么叫 AI 律所What Counts as an AI Law Firm
以 AI 为重要交付基础设施,重新设计人机分工与法律服务交付,并保留明确的专业责任安排。Redesigning human–AI collaboration and legal-service delivery around AI infrastructure while retaining clear professional accountability.
理解这一定义Understanding the definition
本平台将 AI 律所作为研究用语,观察整家机构或明确业务单元的工作方式。它不必是无人律所,也不要求整所同时转型。目录另含早期工具采用与孵化实践,是否改变交付机制应按具体范围和证据判断。AI law firm is a research term here. We examine the operating model of a firm or a defined practice unit. It need not be autonomous or transformed firm-wide. The broader directory also observes early tool adoption and incubation; changes to delivery are assessed within the stated scope and evidence.
补充阅读:工作特征与定义边界Further reading: characteristics and boundaries
01
律师执业与责任Professional responsibility
律师承接法律事务并承担责任,技术与运营可以由不同团队协作支持。Lawyers take on the matter and its responsibility, supported by technology and operations teams.
展开说明Read more
律师执业与责任,以及技术与运营的协作安排Professional accountability and technology–operations collaboration
执业律师与相关法律服务主体承担专业责任,技术和运营可由内部团队或外部伙伴支持。部分实践分别设置律所与科技公司 / MSO,须分别说明专业服务、技术运营及资本安排;主体分设不自动证明合规或转型程度。Licensed professionals and the relevant legal provider retain professional accountability. Technology and operations may be supported internally or externally. Some practices use a law firm and a technology company / MSO; legal, technical and capital arrangements should be stated separately. Separate entities alone establish neither compliance nor transformation depth.
02
AI 进入交付流程AI within delivery
把重复工作交给 AI,律师设定规则、复核例外并作出专业判断。AI handles repeatable work; lawyers set rules, review exceptions and exercise judgment.
展开说明Read more
AI 进入交付流程,人机分工被重新设计AI inside the delivery workflow, with a redefined human–AI split
AI 成为律师的副驾驶,是处理工作的前置程序、工作流中的固定环节;律师从逐字执行转向设定规则、复核例外、把握判断。分工的变化还不止于律所内部——有的律所允许客户直接与 AI 互动、通过 AI 自助完成部分法律工作。AI works as a lawyer's copilot — a fixed step in the workflow that comes before the lawyer does; lawyers shift from drafting every word to setting rules, reviewing exceptions and exercising judgment. The split is changing beyond the firm, too: some firms let clients interact with AI directly and self-serve parts of their legal work.
03
专业与技术协作Legal and technical collaboration
AI、产品与工程能力进入团队,持续建设知识与工作流。AI, product and engineering capabilities help the team build knowledge and workflows.
展开说明Read more
新岗位出现:AI、产品与工程进入律所New roles: AI, product and engineering come in-house
越来越多的律所设有专门的 AI、产品或工程岗位——由内部团队负责工具选型与落地、知识库建设、工作流设计乃至模型微调,把 AI 能力沉淀为律所自身的基础设施,而不是完全依赖外部工具商或外包。A growing number of firms now staff dedicated AI, product and engineering roles — an in-house team owning tool selection and adoption, knowledge-base building, workflow design and even model fine-tuning, turning AI capability into the firm's own infrastructure rather than something fully outsourced to vendors.
04
交付与定价变化Delivery and pricing
固定收费、订阅和产品化服务成为可观察的变化信号。Fixed fees, subscriptions and productized services are signals to observe.
展开说明Read more
定价模式随之转变Pricing shifts with it
从以计时收费为主,转向固定收费、订阅制或产品化的业务单元——把法律服务拆成价格可预期、交付可重复的单元。固定收费既不是必要条件,也不是充分条件,仅作为值得注意的信号,不单独作为判定标准。From billable hours toward fixed fees, subscriptions or productized service units — legal work packaged with predictable prices and repeatable delivery. A telling signal, but never a test on its own — fixed fees are neither necessary nor sufficient.
什么不是 AI 律所What falls outside the definition
以下情形需与直接承接法律服务的组织区分。它们可以进入相关观察池,但不单凭工具采用、平台自称或投资活动认定为 AI 原生法律交付组织。Distinguish the following from organizations directly delivering legal services. They may enter the wider observation pool, but tool adoption, self-description or investment activity alone does not establish an AI-native legal delivery organization.
i
法律科技公司Legal Tech Companies
为律师提供工具而非直接向终端客户提供法律服务。如 Harvey、Clio、Legora——它们是 AI 律所的基础设施供应商。They build tools for lawyers rather than serving end clients directly — e.g. Harvey, Clio, Legora. They are infrastructure providers to AI law firms.
ii
无律师参与的自动化平台Automated Platforms Without Lawyers
无持牌律师监督、无律所实体承接专业责任的自动化服务。如 DoNotPay——绕开了律师监管框架,面临合规争议。Automated services without licensed lawyer supervision or a law firm entity bearing professional liability — e.g. DoNotPay, which bypasses regulatory frameworks and faces compliance challenges.
iii
传统律所的常规数字化Conventional Digitalization
常规工具采用与法律交付重构分别记录。仅采用工具可作为早期转型观察,但需要明确范围;AI 是否改变团队流程、服务交付或组织运营,须另看证据。Routine tool adoption and redesigned legal delivery are recorded separately. Tool use can be observed as early transformation with a clear scope; effects on team processes, delivery or operations need separate evidence.
用六个问题,看懂 AI 律所怎样运作Six questions about how AI law firms operate
六个问题帮助理解机构怎样工作;原有三维分类用于检索与记录,两者不增加新的评分。These questions explain operating mechanisms; the existing taxonomy supports retrieval and recording. They do not add a score.
01
客户买到什么?What does the client buy?
明确事项、持续服务或专家项目分别交付什么,范围与例外怎样说明。Define the output, scope and exceptions of a task, ongoing service or expert project.
看懂一项法律服务中的人机分工Understanding the human–AI division of work
01
需求与事项范围Scope the matter
客户提出需求,律师确认事项、边界与委托安排。The client presents a need; the lawyer confirms scope, boundaries and engagement.
02
规则与任务分流Rules and routing
依据业务背景、模板与风险规则,确定处理路径。Business context, templates and risk rules inform the processing route.
03
AI 处理可重复工作AI processing
在约定流程内整理、分析与起草,形成供复核的成果。AI organizes, analyzes and drafts within the agreed workflow for review.
04
专业判断与交付Judgment and delivery
律师核验、处理例外并作出专业判断,向客户承担责任。Lawyers verify, handle exceptions and make professional judgments with accountability to the client.
05
反馈与持续改进Feedback and improvement
记录使用反馈、返工与异常,经确认后更新规则和流程。Record feedback, rework and exceptions, then update approved rules and workflows.
这是理解分工的示意。具体步骤、授权与复核安排取决于业务和监管要求。This illustrates roles. Actual steps, authorization and review depend on the matter and regulatory requirements.
技术能力可以自建、采购或共建Build, buy or co-develop capabilities
技术与组织能力如何配置Configuring technology and organizational capability
需要配置的能力Capabilities to configure
AI 平台研发与持续迭代,让模型能力跟得上业务变化Building and iterating the AI platform as the business evolves
法律知识库与 RAG 基建,激活律所的历史知识与文书资产Legal knowledge bases and RAG infrastructure that activate a firm's accumulated know-how
把律师的经验流程化、产品化,沉淀为可复用的工作流Turning lawyer expertise into repeatable, productized workflows
数据保护、权限与部署方案,按执业要求设计(私有化/本地化)Data protection, access control and deployment designed to bar requirements (on-premise / local)
效果评估与长期运营,对交付质量持续负责Outcome evaluation and long-term operations, accountable for delivery quality
合作能力的判断要点Assessing collaboration capability
理解法律业务的语义与执业责任边界,不做浅层「技术外包」Understands legal semantics and the boundaries of professional responsibility — no shallow "IT outsourcing"
模型中立,不被单一模型厂商锁定Model-neutral, never locked to a single vendor
工程 × 法律的双语团队,能与合伙人直接对话Bilingual engineering × legal teams that can talk to partners directly
按试点结果决定持续运营或共建,尊重律所数据主权Commits based on pilot results; respects the firm's data sovereignty
技术能力可以自建、采购或与伙伴共建,须结合业务范围、团队能力和成本判断;选择合作伙伴不等于必须设立双主体。Technology can be built in-house, procured, or co-developed with a partner, judged against scope, team capability and cost; choosing a partner does not require a dual structure.
方式Approach
重点判断What to assess
自建Build
内部专业与工程能力、维护责任及持续投入。Internal legal and engineering skills, maintenance ownership and ongoing resources.
采购Buy
现有产品覆盖范围、规则适配、权限与服务支持。Coverage, rule adaptation, access control and service support.
共建Co-develop
双方职责、具体交付、迭代机制与知识使用安排。Responsibilities, concrete delivery, iteration and knowledge-use arrangements.
为什么是现在Why now
AI 改变了部分法律工作的效率;当传统收费主要与耗时挂钩时,效率提升与收入安排可能产生张力,经营方式的变化仍取决于服务范围与交付要求。AI changes the efficiency of some legal work. When traditional fees primarily follow time spent, efficiency and revenue arrangements may diverge; changes still depend on scope and delivery requirements.
2025 年之后的新设与转型实践受到更多关注,可以从以下三类条件理解其出现;不同机构的具体成因仍应分别观察。New ventures and transformation projects have drawn attention since 2025. Three conditions help explain this, while individual causes should be examined separately.
其一,模型能力扩展了任务范围:前沿大模型对交易文书、监管规则的理解与起草,在具体任务、规则和复核条件下逐步具备应用价值。此前的法律科技已包括检索、标记、文档自动化与流程协作,新的变化在于能力怎样进入完整交付。First, model capability expands the task range: interpreting and drafting transactional documents and regulatory material can have value under task-specific rules and review. Earlier legal technology already included search, tagging, document automation and workflow collaboration. The question is how new capabilities enter complete delivery.
其二,成本结构提供新的探索空间:部分顶级律所合伙人每小时上千美元的费率受到客户关注,合同审查、基金设立文件和合规问卷等高重复工作也存在不同服务价格。AI 可能改变重复环节的投入,但不能据此统一推断边际成本降低一个数量级;固定收费与订阅并非首次出现,其适用性仍需结合专业复核、例外处理与全部运营成本验证。Second, costs create room for experimentation: clients scrutinize high partner rates, while repetitive contract review, fund documents and compliance questionnaires have varied prices. AI may change effort in repeatable steps, but no universal order-of-magnitude reduction follows. Flat fees and subscriptions already existed; suitability depends on review, exceptions and full operating costs.
其三,资本与制度环境影响组织选择:部分实践通过 MSO 等结构探索资本与运营参与方式,英国 ABS、美国亚利桑那 ABS 与犹他州沙盒分别提供当地条件下的制度或试验框架,黑石等资管机构也以投资者与客户身份参与。具体主体、辖区、许可状态与专业责任仍需分别核对,不能由某种结构名称直接推断合规。Third, capital and institutions shape organizational choices: some practices explore investment and operations through MSOs, while UK ABS, Arizona ABS and the Utah sandbox provide distinct local frameworks. Asset managers such as Blackstone also participate as investors and clients. Entities, jurisdiction, authorization and accountability require separate checks; a structure name does not establish compliance.
何者科技搭建这个平台,正是基于对这轮变化的判断:法律服务供给方的工作方式正在发生结构性重构,新项目与新模式的涌现不是营销噪音,而是上述三条曲线交汇的产物。因此我们以中文视角持续追踪全球实践——收录项目、记录关键事件、以三个维度刻画工作方式并核验证据,服务企业法务部与律师事务所的双端决策,也为研究者与投资者提供系统性参考。HEZHER built this platform on exactly that judgment: the way legal services are produced is being structurally rebuilt, and the wave of new ventures is not marketing noise but the product of those three curves converging. We therefore track global practice from a Chinese perspective — indexing projects, recording key events, profiling work practices along three dimensions with verified evidence — serving decision-making on both the corporate-legal and law-firm sides, and offering researchers and investors a systematic reference.
关键里程碑Key Milestones
以下只收录行业层面的分水岭事件——监管为新模式开门、责任承保、法律实体形态创新、顶级机构亲自下场;单个律所的上线、融资数据或营销式报道不在此列。Only industry-level watershed events are listed here — regulators opening the door, liability insurance, new entity forms, and top institutions stepping in themselves; individual launches, funding figures and PR pieces are not.
2025.05
监管授权 · Garfield 获英国 SRA 授权。2025 年 5 月公告将其称为首家获 SRA 授权、通过 AI 提供受监管法律服务的律所,范围为英格兰及威尔士。公告明确客户逐步批准、监督监控及指定律师最终负责;该事件展示新的交付与责任安排,没有豁免执业责任。Authorization · In May 2025, the SRA described Garfield as its first authorized law firm delivering regulated services through AI in England and Wales. The notice specifies client approval at each step, supervision and ultimate accountability of named solicitors; it does not waive professional responsibility.
相关制度探索 · 特拉华州与 Norm Ai 提出 AIC。2026 年 7 月联合署名文章讨论由 AI 管理日常事务的公司形态与拟议监管沙盒,包括主体、资本、记录与责任安排。这是对 AI 参与企业经营的新主体形式的探索,不直接等于 AI 获准法律执业;提案、试点与生效状态分别记录。Institutional exploration · A July 2026 joint article on Delaware and Norm Ai discusses an AI-managed company form and a proposed sandbox, including entity, capital, record-keeping and responsibility. This explores AI-managed business, not authorization to practise law; proposal, pilot and effective status are distinct.
责任保险 · Crosby 探索为 AI agents 的法律责任投保。2026 年 7 月报道引述其 CEO 的计划,并说明当时尚未披露保险公司。AI 参与交付后的承保和责任安排值得持续观察;相关探索也见于 Orbital、Corgi 等案例,但投保计划不等于保单生效或自主执业许可。Liability insurance · A July 2026 report quotes Crosby’s CEO on plans to insure AI-agent liability and notes that no insurer had been named. Cover and responsibility as AI enters delivery remain worth observing, alongside Orbital and Corgi examples. Plans are not evidence of effective policies or permission for autonomous practice.
资本与客户关系 · Bloomberg Law 于 2026 年 8 月报道,Blackstone、Coatue、Bain Capital 与 Khosla 既投资 Norm Ai,又是 Norm Law 具名客户。报道引述 Blackstone 称,部分投资事项由 Norm Law 处理,相较传统供应商已有实际节省,但未披露幅度,且当时尚未实质影响其外部律所使用。此前 Norm Ai 公布 1.2 亿美元 C 轮,由 Khosla 领投,Vanguard、New York Life、TIAA 等参投。融资主体与法律服务提供方分别记录。Capital and client relationships · Bloomberg Law reported in August 2026 that Blackstone, Coatue, Bain Capital and Khosla both backed Norm Ai and were named Norm Law clients. Blackstone reported savings on certain matters without a percentage, and no material effect on outside-firm use at that time. Norm Ai had announced a $120 million Series C led by Khosla, with Vanguard, New York Life, TIAA and others participating. The financing entity and legal provider are distinguished.
三维分类:研究工作方式,而非评选优劣Three dimensions: work practices, not rankings
本目录用三个相互独立的维度描述每一个观察对象。先确定观察范围——整家机构、某条业务线、创新部门,或平台与律所的组合——再从三个维度分别判断。The Index describes each observed unit along three independent dimensions. Fix the scope first — a whole firm, a service line, an innovation unit, or a platform–firm combination — then assess each dimension separately.
01
转型深度Transformation depth
AI 改变了多少工作方式?How much has AI changed the way work is done?
工具辅助Tool-assisted辅助个人任务Assists individual tasks
流程整合Workflow-integrated进入团队可复用流程Enters repeatable team workflows
服务重构Service-redesigned重设计分工与交付Reshapes roles and delivery
组织原生Organization-native形成知识与改进机制Builds knowledge and improvement mechanisms
待核验:资料不足时单独标注。转型深度不代表服务质量或推荐排名。Unassessed: insufficient evidence is marked separately. Depth is not a ranking of service quality or recommendations.
明确任务或成果单元 / 一段时间内承接约定工作 / 围绕交易、争议、复杂事项组织专家A defined task or output unit / agreed work handled over time / specialists organized around deals, disputes or complex matters
组织来源Organizational origin
由谁发起、平台与专业如何耦合?Who initiated it, and how are platform and expertise coupled?
看发起团队、既有业务是否延续、平台与专业供给的关系The initiating team, continuity of existing practice, and the platform–professional supply relationship
判断流程:逐项读取档案 → 确认观察对象与业务范围 → 为三个维度分别记录判断与理由 → 回查关键或冲突的原始来源 → 展示信息缺口与核验日期。三个维度不能相互替代,也不代表认证或推荐。Assessment flow: read each archive → fix the observed unit and scope → record a rationale per dimension → re-check primary sources for decisive or conflicting claims → disclose gaps and review dates. The dimensions are not interchangeable, and inclusion is not certification or endorsement.
目录页在这些维度之外,还可按所在地区(北美 / 欧洲 / 拉美 / 亚太 / 中东与非洲)与业务领域(合同、诉讼、知识产权、合规等)交叉筛选。来源与核验记录说明具体事实的支持情况,不对项目作整体评级,也不参与转型深度的判定。Beyond these dimensions, the directory also supports cross-filtering by region (North America / Europe / Latin America / Asia Pacific / Middle East & Africa) and practice area (contracts, litigation, IP, compliance and more). Source and review records describe support for specific claims, not an overall project rating, and play no part in assessing transformation depth.
研究边界与核验说明Research boundaries & verification
为保证口径一致,以下情形本目录一律不作推断。它们是对上面三个维度的补充约束,而非新的分类。To keep the standard consistent, the Index makes none of the following inferences. These are constraints on the three dimensions above, not additional categories.
工作方式、专业质量与法律责任分别判断;资料不足时展示缺口。Work practices, professional quality and legal responsibility are assessed separately. Evidence gaps are disclosed.
查看六项研究说明Read the six research notes
i
深度 ≠ 质量或排名Depth ≠ quality or ranking
转型深度不是服务质量、盈利能力或推荐排名。组织原生须同时有常态化分工、知识积累与持续改进,不能仅凭自称、融资或自研平台认定;保留人工专业判断也不代表落后。Depth is not a ranking of quality or profitability. Organization-native requires standing roles, accumulated knowledge and improvement loops — never self-description, funding or a proprietary platform alone. Retaining human judgment is not backwardness.
ii
标签不能互推Labels don't imply each other
AI 原生不等于无人化;固定收费既非必要也非充分条件。局部业务线不代表整所;服务路线可并存;组织来源不等于法律主体或监管资格;顶级律所属既有律所背景,不与「AI 原生」并列。Native is not autonomy; flat fees are neither necessary nor sufficient. A service line is not the whole firm; routes may overlap; origin is not legal entity or regulatory status; top firms are an established-firm background, not an alternative to "AI-native".
iii
主体与责任单独核验Entities and liability verified separately
分别确认谁签委托、谁出法律意见、谁供技术与运营、谁担责。单/双主体、收购、ABS 是不同属性,不混为一类;客户数据不因平台归属被推定为某方自有;个人律师资格也替代不了机构监管记录。Identify separately who signs the engagement, gives legal advice, supplies technology and operations, and bears responsibility. Entity count, acquisitions and ABS are distinct attributes; platform ownership does not establish client-data ownership; a lawyer's personal qualification is not an organization's regulatory record.
iv
缺口 ≠ 落后Gaps ≠ low maturity
资料不足以判断时记为「待核验」,不降格为工具辅助、也不暗示转型程度低。阶段零(纯人工流程)仅作比较基线,不作为本轮收录对象。When evidence is insufficient, an entry is marked Unassessed — never downgraded to tool-assisted nor read as low maturity. Stage zero (fully manual work) is a baseline only, not a target of this round.
v
收录边界Inclusion scope
本目录现收录约 112 个项目:收录传统律所的 AI 创新部门,但不收录未开展 AI 创新的传统律所。纯工具采用与孵化投资可作观察对象,但须如实注明范围,不冒充已重构的法律服务。收录仅表示列为研究观察对象;具体事实的来源与核验情况分别记录,不构成对机构的整体评级、认可、认证或推荐。The directory now covers ~112 projects, including AI innovation units of conventional firms, while excluding conventional firms without AI innovation. Tool adoption and incubation can be observed but their scope is stated plainly, never passed off as redesigned delivery. Inclusion identifies a subject of research; sources and review status are recorded for specific claims, without an overall organization rating, endorsement, certification or recommendation.
vi
数据口径Data basis
统计以「项目数」计,项目数不等于独立机构数;关联的平台与律所可能分别观察。旧链接保留历史分类视图,不自动换算为新阶段。完整声明见页脚。Counts are of projects, not independent organizations; linked platforms and firms may be tracked separately. Legacy links keep a historical-category view without auto-mapping to a new stage. Full disclaimer in the footer.
资料状态说明Information-status definitions
未披露Not disclosed
已收录的公开材料未给出该项信息。The reviewed public material does not state the item.
尚未找到资料No source located
尚未取得可核对的支持材料。Checkable supporting material has not been obtained.
不适用Not applicable
该字段不属于本条观察对象的范围。The field falls outside this record’s observed scope.
研究覆盖与持续更新Research coverage and updates
以下为当前档案统计。项目数不等于独立机构数;融资口径可能包含同组技术平台,具体方法见目录数据分析。These figures describe the current archive. Project counts are not independent-entity counts; funding may include associated technology platforms. See the directory methodology.
项目数不等于独立机构数。融资按档案中的融资主体去重,优先采用累计披露;仅美元金额纳入合计,内部预算、关联项目重复记录及未换算的外币金额不计入。关联平台的融资不等于律所融资,近似金额保留原始精度限制。国家查询按档案单一主要所在地归一化,多地项目单列,不自动划为某一国家。创始人背景和目标市场按档案明确描述归纳,不以 CEO 或扩张计划推断技术背景或当前覆盖。Projects are not independent organizations. Funding is deduplicated by financing entity, preferring cumulative disclosures. Only USD amounts enter the total; internal budgets, duplicate related records and unconverted currencies are excluded. Platform funding is not law-firm funding, and approximate inputs limit precision. Country queries normalize a single recorded primary location; multi-location records remain separate. Background and target-market summaries follow explicit archive descriptions, not CEO titles or expansion plans.