曼昆周报 | Ai×Web3 每周法律简报

AI & Web3 Law — Key Developments

⚖ U.S. COURTS

June 3 (ongoing) — CNN v. Perplexity: First major broadcast-news AI copyright suit proceeds. CNN filed suit against Perplexity on May 29, alleging its AI search product systematically scrapes and reproduces CNN’s copyrighted news content without authorisation or payment. This case proceeded into the June 3 week with discovery orders expected. CNN becomes the first major U.S. broadcast-news organisation to litigate rather than license, signalling a hardening stance across legacy media. Follows the landmark $1.5 billion Bartz v. Anthropic settlement (pending final approval) and the $3.1 billion Universal Music/Concord/ABKCO v. Anthropic suit filed in January 2026.
June 3–10 — OpenAI MDL 3143 discovery: 88 million ChatGPT logs ordered produced. Multiple copyright lawsuits against OpenAI were consolidated into one big case called In re OpenAI Copyright Litigation (N.D. Cal.), the court previously ordered OpenAI to produce 20 million logs, then expanded in March 2026 to 78 million + 10 million additional logs. Discovery activity continued through this week, with plaintiffs’ counsel reviewing logs for evidence of memorisation and reproduction of copyrighted text. The scale of production is unprecedented in AI IP litigation.
June 9 (breaking) — Taiwan reports criminal chip export penalty bill. Bloomberg reported that Taiwan is considering legislation imposing criminal penalties for the unauthorised export of advanced AI chips (Nvidia Blackwell-class) to all of China — not just blacklisted firms such as Huawei. TSMC shares fell on the news. Taiwan has agreed “directionally” to align with U.S. rules but the final scope is still being negotiated. If enacted, this would be the most expansive extension of the AI chip control regime outside the United States.

U.S. LEGISLATION — FEDERAL

June 2 — Trump signs AI Executive Order on Innovation and Security. The most significant U.S. federal AI directive of 2026. Key provisions: (1) AI developers may voluntarily give the government up to 30 days early access to frontier models before public release; (2) agencies have 60 days to establish a classified benchmark designating “covered frontier models”; (3) a new AI Cybersecurity Clearinghouse enables voluntary public-private vulnerability identification; (4) the DOJ is directed to prioritise criminal enforcement against AI-assisted computer intrusion; and (5) mandatory government pre-clearance or licensing for AI development is expressly prohibited.
June 1–2 — CLARITY Act placed on Senate calendar; floor vote imminent. The Digital Asset Market Clarity Act was placed on the Senate Legislative Calendar (General Orders No. 423) on June 1, after clearing the Senate Banking Committee 15–9 on May 14. Would grant the CFTC exclusive jurisdiction over digital commodity spot markets while preserving SEC jurisdiction over investment-contract assets. Key obstacles: 60-vote floor threshold, ethics provision on government officials’ crypto holdings, and Senate calendar pressure before summer recess. White House aspirational signing date: July 4.
June 1 — Commerce Dept acknowledges AI chip enforcement gap. Commerce issued clarifying guidance reinstating the ban on advanced AI chip exports to China-headquartered companies operating outside China, while acknowledging a potential enforcement lapse that may have allowed hundreds of thousands of Nvidia Blackwell GPUs to flow to Chinese firm subsidiaries in Malaysia, Singapore, and other third countries for over a year. Material compliance exposure for U.S. chip suppliers and their distributors.

U.S. LEGISLATION — STATE

June 4–5 — New York passes historic 3-bill AI package; adjourns sine die. Three landmark bills passed before adjournment on June 5: (1) NY FAIR News Act — requires news organisations to disclose AI-generated content; (2) AI Training Data Transparency Act (A 6578, Senate 54–6 on June 4) — requires generative AI developers to publish a dataset summary for any model offered in New York; (3) Kids AI Chatbot Safety Bill (S 9051, unanimous 137–0 and 60–0) — prohibits companion chatbots for minors. Governor Hochul expected to sign all three.
June 5 (update) — 78 chatbot bills alive in 27 states; California’s 30 AI bills move forward. The Transparency Coalition’s June 5 update reported 78 active chatbot bills in 27 states. Colorado Governor Polis signed four of five 2026 AI bills (original SB 24-205 replaced by narrower SB 26-189, effective January 1, 2027). California’s 30 remaining AI bills are in second-chamber committees ahead of a July 2 recess; sine die is August 31.

CHINA

June 3–10 (ongoing) — 2026 Qinglang AI enforcement campaign active. The CAC’s 2026 Qinglang (清朗) enforcement campaign — coordinated with the Ministry of Public Security — is actively targeting: AI-enabled deepfake fraud and voice-cloning scams; generative AI services without CAC registration; and AI-generated content violating privacy and IP rights. The 2025 campaign resulted in 3,500+ AI products taken down and 960,000+ pieces of illegal content removed.
July 15 deadline approaching (35 days) — CAC Anthropomorphic AI Services Interim Measures. The CAC’s Interim Measures for the Management of Anthropomorphic AI Interactive Services (issued April 10, 2026) take effect July 15, 2026. Govern AI companions, chatbots, and virtual assistants simulating human personality. Key obligations: privacy compliance, safety testing, mandatory AI labelling, and a ban on companion features for minors. No grace period.
Implementation week — CAC/NDRC/MIIT AI Agents Framework enters effect. The joint Implementation Opinions on the Standardised Application and Innovative Development of Intelligent Agents (issued May 8, 2026) are in active implementation. This is the first Chinese regulatory framework treating AI agents as a distinct category from generative AI or LLMs. Risk-tiered structure: healthcare, finance, transportation, judicial services, and public security face mandatory filing and joint-regulator recall mechanisms; lower-risk consumer/productivity applications face lighter-touch self-assessment.

EU

June 1 — EU AI Office appoints 60-expert Scientific Panel; GPAI enforcement live. The European Commission formally appointed the EU AI Act’s 60-member Scientific Panel and Advisory Forum, giving the European AI Office the independent expert infrastructure needed to enforce GPAI model obligations. Enforcement is now operationally active.
June 3 — AI Act transparency consultation closes; Code of Practice on AI labelling finalised. Public consultation on EU AI Act Article 50 watermarking/transparency guidelines closed June 3. The Commission published the final Code of Practice on Marking and Labelling of AI-Generated Content — two months ahead of the August 2 deadline. Standardised visual “AI” label required on all AI-generated text, images, audio, and video. Safe harbour for signatories. New prohibited practices confirmed effective December 2, 2026: AI generating non-consensual intimate imagery or CSAM, with fines up to €35 million or 7% of global turnover.

CALL TO COMPLIANCE ACTIONS

What AI Companies Should Do Now

U.S. — Federal Level

Trump AI EO — Frontier Model Developers: Assess exposure to the voluntary pre-release access framework. Document which models may qualify as “covered frontier models” once the classified benchmarking threshold is published (expected within 60 days of June 2). Engage government affairs counsel on cybersecurity clearinghouse participation.
AI-Assisted Cybercrime Exposure: The EO directs DOJ to prioritise criminal enforcement against AI-assisted unauthorised computer access. Audit AI tools that interact with external systems or automate access-related tasks for inadvertent CFAA exposure.
Training Data Provenance — Ongoing Obligation: Implement or refresh a documented training data due-diligence programme. Any dataset from piracy sources, unlicensed scraping of paywalled content, or content with AI-use restrictions must be identified and quarantined. The $1.5 billion Bartz v. Anthropic settlement is the current damages benchmark.

U.S. — New York

AI Training Data Transparency Act (A 6578): If you are a generative AI developer with any New York users or operations, prepare a public-facing dataset summary disclosure covering training data categories and sources. Build disclosure infrastructure now, before Governor Hochul’s signature makes the obligation imminent.
Kids AI Chatbot Safety Bill (S 9051): If your company operates any AI companion, social, or interactive chatbot accessible to persons under 18 in New York, plan immediate service restrictions or age-verification. The bill passed unanimously — signature is near-certain.
NY FAIR News Act: If your company or client publishes AI-generated news content distributed in New York, implement a disclosure/labelling system for AI-generated articles now.

U.S. — California

30 AI Bills in Second-Chamber Committees: California’s remaining 30 AI bills cover AI liability, biometric data, automated decision systems, and AI in healthcare. Retain state legislative counsel to monitor through the August 31 sine die date.

China

Anthropomorphic AI Services — July 15 Deadline (35 Days): Complete a gap analysis against the CAC Interim Measures immediately. Key obligations: privacy compliance, safety testing, mandatory AI labelling, and a ban on companion features for minors. There is no grace period.
AI Agents Framework — High-Risk Sector Filing: If your client deploys AI agent products in healthcare, finance, transportation, judicial services, or public security in China, begin mandatory filing with the relevant sectoral authority (NHSA, PBOC/NFRA, etc.) in coordination with the CAC.
Qinglang 2026 — Registration Audit: Verify that all generative AI services deployed in China hold current CAC registration. Unregistered services are the primary enforcement target. File for registration or re-registration before the campaign intensifies.

EU

AI Act Transparency Code of Practice — Sign On Now: Signing on to the Code of Practice on AI-Generated Content Labelling is the fastest path to Article 50 compliance before the August 2, 2026 deadline. Implement the standardised visual “AI” label on all AI-generated outputs.
GPAI Model Compliance — Scientific Panel Active: With 60 independent experts now appointed to the EU AI Office Scientific Panel, GPAI enforcement is no longer theoretical. Ensure GPAI model documentation, capability evaluations, and systemic risk assessments are complete and current.

What Web3 Companies Should Do Now

U.S.

CLARITY Act — Token Portfolio Mapping (Urgent): Map your entire token portfolio against the Act’s CFTC/SEC jurisdictional divide now. The decentralisation threshold is the critical variable. Tokens that are “sufficiently decentralised” fall to CFTC commodity jurisdiction; tokens that remain investment contracts stay with the SEC. Engage digital assets counsel immediately.
SEC-CFTC Joint Token Taxonomy — Update Classifications: Review all current token classifications against the March 2026 joint SEC-CFTC Interpretive Release. The taxonomy covers digital commodities, digital securities, digital collectibles, stablecoins, and the treatment of airdrops, protocol staking, and asset wrapping.
GENIUS Act — Stablecoin Reserve Compliance: If you issue or plan to issue a stablecoin in the U.S., confirm 100% reserve backing with qualifying liquid assets per OCC implementing regulations (effective March 2, 2026). Implement reserve auditing and public attestation.
AI Chip Export Controls — Supply Chain Review: If your infrastructure operations involve hardware procurement through third-country intermediaries (Malaysia, Singapore, UAE), conduct an immediate review of all chip sourcing chains. Commerce’s re-clarification makes such arrangements presumptively non-compliant.

Hong Kong

First Stablecoin Licenses Imminent — HKMA: The HKMA confirmed the first batch of stablecoin issuer licenses will be granted “next month.” Ensure reserve documentation, governance structures, and AML/KYC programmes are current. If not yet applied, assess eligibility now.
LEAP Framework — Licensing Developments for Virtual Asset Service Providers: Hong Kong is continuing to refine its virtual asset regulatory regime. The LEAP Framework signals potential future licensing requirements for certain virtual asset-related activities, including trading and custody services. Firms operating or planning to operate in Hong Kong should monitor legislative developments and review their compliance, governance, and risk management frameworks accordingly.

UK

FCA Cryptoasset Perimeter — DeFi and Wallet Exposure: Final guidance expected September 2026. The FCA’s broad interpretation of “arranging” activities could sweep in DeFi front-ends, aggregators, and non-custodial wallets. Obtain a preliminary legal opinion on UK licensing exposure before September.

China

Businesses should continue to monitor regulatory developments relating to digital assets, blockchain technologies and other emerging technology applications, and assess the potential compliance implications based on their specific business activities and operational arrangements.

This briefing is prepared for informational purposes only and does not constitute legal advice.
If you are seeking AI legal governance, or Web3 & digital assets compliance, feel free to reach out to our lawyers.

Many thanks!

Mankun Law PLLC

⚡ 本周回顾：2026年6月3日至10日

近期，全球AI与Web3监管持续升温。美国围绕AI版权、前沿模型监管及数字资产立法动作频频；欧盟《AI法》进入实质执法阶段；中国则加快推进拟人化AI、智能体及生成式AI服务监管框架落地。与此同时，数字资产市场监管规则也正在逐步明晰。

本期周刊梳理过去一周AI与Web3领域最值得关注的监管、立法及司法动态，并结合最新趋势提出企业合规建议，供读者参考。

AI 与 Web3 法律——重要动态

美国法院

6月3日（持续进行中）——CNN诉Perplexity案：首起重大广播新闻AI版权诉讼推进。 CNN于5月29日提起诉讼，指控Perplexity的AI搜索产品系统性地抓取并复制CNN受版权保护的新闻内容，且未获授权、未支付任何费用。该案于6月3日当周进入审前证据开示程序。CNN成为美国首家选择诉诸法庭而非授权合作的主流广播新闻机构，标志着传统媒体对AI公司的立场趋于强硬。此案紧随Bartz诉Anthropic案15亿美元和解协议（终止书面批准程序尚待完成）及2026年1月提起的环球音乐/Concord/ABKCO诉Anthropic 31亿美元索赔案之后。

6月3日至10日——OpenAI MDL 3143证据开示：法院命令提交8800万条ChatGPT日志。多起针对OpenAI的版权诉讼已合并为加州北区联邦法院受理的OpenAI版权诉讼案（In re OpenAI Copyright Litigation）。法院此前命令OpenAI提交2000万条日志，并于2026年3月扩展至7800万条另加1000万条附加日志。本周证据开示活动持续推进，原告律师正审查日志，寻找ChatGPT记忆并复制版权文本的证据。此次证据开示规模在AI知识产权诉讼史上史无前例。

6月9日（突发）——台湾拟出台AI芯片出口刑事处罚法案。 彭博社6月9日报道，台湾正考虑立法，对向中国大陆（包括未被列入黑名单的企业，不仅限于华为等）未经授权出口先进AI芯片（英伟达Blackwell级别）的行为处以刑事处罚。消息传出后，台积电股价应声下跌。台湾已原则上同意与美国规则接轨，但最终范围仍在磋商中。若获通过，将是美国境外AI芯片管制体系最大范围的延伸。

美国立法——联邦层面

6月2日——特朗普签署《促进先进人工智能创新与安全》行政令。 这是2026年最重要的美国联邦AI政策指令。主要条款：（1）AI开发者可自愿在公开发布前向政府提供最长30天的前沿模型早期访问权；（2）各机构须在60天内建立保密基准，用于认定受涵盖前沿模型；（3）新建AI网络安全信息共享中心，实现政府与行业之间的自愿漏洞识别协作；（4）司法部被指令优先追诉利用AI实施未授权计算机入侵的犯罪行为；（5）明确禁止政府对AI开发设立强制性审批或许可要求。

6月1日至2日——《数字资产市场明晰法案》列入参议院议程，全体表决在即。《数字资产市场明晰法案》（CLARITY Act）于6月1日列入参议院立法日历（一般命令第423号），此前该法案于5月14日以15票对9票通过参议院银行委员会审议。该法案将赋予商品期货交易委员会（CFTC）对数字商品现货市场的专属管辖权，同时保留证券交易委员会（SEC）对投资合同类资产的管辖权。主要障碍：需60票克服冗长辩论程序、涉及政府官员持有加密资产的道德条款，以及暑假前参议院日程压力。白宫期望在7月4日前签署生效。

6月1日——美国商务部承认AI芯片出口管制存在执法漏洞。 商务部发布澄清指导意见，重新恢复对向在华以外地区运营的中资企业出口先进AI芯片的禁令，同时承认此前可能存在的执法空白——大量英伟达Blackwell GPU或已通过马来西亚、新加坡等第三国渠道流入中资企业子公司，持续时间超过一年。此举对美国芯片供应商及其经销商构成重大合规风险。

美国立法——州层面

6月4日至5日——纽约州通过历史性三项AI法案；立法会期届满休会。 6月5日休会前，三项重要法案获得通过：（1）《纽约州AI新闻公平法》——要求新闻机构披露实质性或全部由AI生成的内容；（2）《AI训练数据透明度法》（A 6578，6月4日参议院54票对6票通过）——要求生成式AI开发者公布其在纽约提供的任何模型所用训练数据集摘要；（3）《儿童AI聊天机器人安全法案》（S 9051，众议院137票对0票、参议院60票对0票全票通过）——禁止AI公司向未成年人提供陪伴型聊天机器人。州长霍楚尔预计将全部签署。

6月5日（最新动态）——全美27个州存在78项聊天机器人法案；加利福尼亚州30项AI法案持续推进。 透明度联盟6月5日立法简报显示，全美27个州共有78项聊天机器人相关法案处于审议中。科罗拉多州州长Polis已签署2026年五项AI法案中的四项（原SB 24-205已被范围更窄的SB 26-189取代，后者自2027年1月1日起生效）。加利福尼亚州剩余30项AI法案正在第二议院委员会审议，7月2日夏季休会，本届会期届满日为8月31日。

中国

6月3日至10日（持续进行）——2026年清朗AI专项整治行动持续推进。 国家互联网信息办公室2026年度清朗专项行动与公安部协同开展，重点打击：AI深度伪造（Deepfake）诈骗和声音克隆冒充；未经网信办备案的生成式AI服务；以及侵犯隐私权和知识产权的AI生成内容。2025年度行动共下架3500余款AI产品、清除96万余条违法内容。

7月15日截止（距今35天）——网信办《拟人化AI互动服务管理暂行办法》合规期限临近。 网信办《拟人化人工智能互动服务管理暂行办法》（2026年4月10日发布）将于2026年7月15日正式施行。适用于模拟人类个性的AI伴侣、聊天机器人及虚拟助手服务。主要义务：隐私合规、安全测试、强制AI标识，以及禁止向未成年人提供伴侣功能。不设过渡期。

实施周——网信办/发改委/工信部《智能体应用规范与创新发展实施意见》进入落地阶段。 《关于规范应用和创新发展智能体的实施意见》（2026年5月8日联合发布）本周进入正式落地阶段。这是中国首个将AI智能体作为独立监管类别（有别于生成式AI或大语言模型）进行规范的监管框架。实行风险分级管理：医疗健康、金融、交通运输、司法、公共安全等高风险领域须履行强制备案和联合监管机构召回机制；低风险消费和生产力类应用适用轻量化自我评估管理。

欧盟

6月1日——欧盟AI办公室任命60名专家科学小组；通用人工智能（GPAI）执法正式启动。 欧盟委员会正式任命《欧盟AI法》科学小组及咨询论坛的60名成员，为欧盟AI办公室提供落实通用人工智能（GPAI）模型合规执法所需的独立专家基础设施。GPAI条款执法现已全面启动。

6月3日——《AI法》透明度磋商截止；AI生成内容标识实践准则正式发布。 欧盟《AI法》第50条水印/透明度指南公众磋商于6月3日截止。欧盟委员会随即发布最终版《AI生成内容标识实践准则》，较8月2日透明度规则生效节点提前两个月。主要要求：所有AI生成的文本、图像、音频及视频须附标准化AI视觉标识；签署方可获安全港保护。新增禁止行为自2026年12月2日起生效：利用AI生成非自愿亲密影像或儿童性剥削材料（CSAM），最高处以3500万欧元或全球营业额7%的罚款。

合规行动建议

AI 企业当前应采取的行动

美国——联邦层面

特朗普AI行政令——前沿模型开发者：评估自愿预发布访问框架下的风险敞口。在保密基准阈值公布后（预计6月2日起60天内），记录哪些模型可能被认定为受涵盖前沿模型。就参与网络安全信息共享中心事宜咨询政府事务顾问。AI辅助网络犯罪风险：该行政令指示司法部优先追诉AI辅助未授权计算机访问犯罪。对与外部系统交互或自动化访问相关任务的AI工具进行审计，排查无意中触犯《计算机欺诈和滥用法》（CFAA）的风险。训练数据来源合规——持续性义务：建立或更新有文件记录的训练数据尽职调查程序。来源于盗版网站、对付费内容的未授权抓取、或含有禁止AI使用限制的数据集，须立即识别并隔离。Bartz诉Anthropic案15亿美元和解金额是当前的损害赔偿参考基准。

美国——纽约州

《AI训练数据透明度法》（A 6578）：如您是在纽约拥有用户或业务的生成式AI开发者，请准备面向公众的训练数据集摘要披露文件，涵盖训练数据的类别和来源。在霍楚尔州长签署法案前，应立即着手构建披露基础设施。《儿童AI聊天机器人安全法案》（S 9051）：如您的公司在纽约运营任何可供18岁以下用户访问的AI伴侣、社交或互动聊天机器人产品，应立即规划服务限制或年龄验证措施。该法案全票通过，州长签署几乎板上钉钉。《纽约州AI新闻公平法》：如您的公司或客户在纽约发布AI生成新闻内容，应立即为AI生成文章部署披露与标识系统。

美国——加利福尼亚州

30项AI法案进入第二议院委员会审议：加利福尼亚州剩余30项AI法案涵盖AI责任、生物特征数据、自动决策系统及医疗AI等领域。应聘请州级立法顾问，持续跟踪至8月31日届满日，评估获通过法案的合规义务。

中国

拟人化AI服务——7月15日截止（剩余35天）：立即对照网信办《暂行办法》开展差距分析。主要义务：隐私合规、安全测试、强制AI标识，以及禁止向未成年人提供伴侣功能。不设过渡期，不得心存侥幸。智能体框架——高风险行业备案：如您的客户在中国医疗健康、金融、交通运输、司法或公共安全领域部署AI智能体产品，应立即配合网信办启动向相关行业主管部门（医疗：国家卫健委；金融：中国人民银行/国家金融监督管理总局等）的强制备案程序。2026年清朗行动——备案情况自查：核实所有在华运营的生成式AI服务是否持有有效的网信办备案。未备案服务是本次专项行动的重点打击对象。在行动力度加强前，尽快完成备案或重新备案申请。

欧盟

欧盟AI法透明度实践准则——立即签署：签署《AI生成内容标识实践准则》是在2026年8月2日截止日前实现第50条合规的最快路径。对所有AI生成内容输出实施标准化AI视觉标识。GPAI模型合规——科学小组已正式运作：随着60名独立专家加入欧盟AI办公室科学小组，GPAI执法已不再停留于理论层面。确保GPAI模型文档、能力评估及系统性风险评估完整且处于最新状态。

Web3 企业当前应采取的行动

美国

《明晰法案》——代币组合分类梳理（紧急）：立即对照该法案CFTC/SEC管辖权划分，对全部代币组合进行分类梳理。去中心化程度认定是关键变量——充分去中心化的代币归属CFTC商品管辖；仍具投资合同性质的代币归属SEC管辖。立即聘请数字资产专项律师。SEC-CFTC联合代币分类框架——更新分类：对照2026年3月SEC-CFTC联合解释性声明审查所有当前代币分类。分类框架涵盖数字商品、数字证券、数字收藏品、稳定币，以及空投、协议质押和资产包装的处理规则。《GENIUS法案》——稳定币储备合规：如您在美国发行或计划发行稳定币，根据货币监理署（OCC）实施法规（2026年3月2日起生效），确认以合格流动资产100%全额背书。建立储备审计和公开证明机制。AI芯片出口管制——供应链合规审查：如您的基础设施运营涉及通过第三国中间商（马来西亚、新加坡、阿联酋）采购硬件，应立即全面审查所有芯片采购链。商务部的重新澄清使此类安排被推定为违规。

香港

首批稳定币牌照在即——香港金管局：金管局已确认首批稳定币发行人牌照将于下月颁发。确保储备文件、治理架构及反洗钱/了解你的客户（AML/KYC）合规程序处于最新状态。如尚未申请，现在应立即评估申请资格。LEAP框架——虚拟资产交易商及托管机构牌照法案即将推出：香港LEAP框架预示今年将通过立法建立虚拟资产交易商和托管机构牌照制度。构建香港注册实体时，须与中国内地业务之间设立清晰的运营和法律防火墙。

英国

英国FCA加密资产监管范围——DeFi及钱包合规风险：最终指引预计2026年9月发布。FCA对安排行为的宽泛解释可能将DeFi前端界面、聚合器及非托管钱包纳入英国加密资产牌照管理范围。在9月前，应就英国牌照合规风险取得初步法律意见书。

中国

着重规避—内地业务：任何与中国内地有关联的客户或产品，必须着重避免加密货币交易、稳定币发行、现实世界资产（RWA）代币化及Web3服务提供。香港架构必须维持清晰的运营和法律防火墙。

本简报仅供参考，不构成法律意见。如需AI法律治理或Web3与数字资产合规服务，欢迎联系我们的律师。感谢您的阅读！

曼昆周报 | Ai×Web3 每周法律简报