本周回顾：2026年7月30日 — 8月5日本周三个关键判断
对美国AI企业而言：训练数据的来源，如今已是核心业务风险，而非技术细节
美国法院正在审理一批关于AI模型如何训练的版权案件，本周欧洲一家法院更以判决强化了这一趋势——认定音乐应用Suno必须为其训练所用的歌曲付费获得授权，且训练与生成两端都要付费。对任何一家开发AI的美国企业而言，信息都是一致的：如果你无法证明训练数据是合法取得的，这就是一项可能带来真金白银损失的业务风险。能够清楚记录数据来源的企业，将比做不到的企业得到更好的保护。
在中国，AI陪伴与拟人化产品已开始被处罚，且执法节奏很快
中国针对拟人化与AI陪伴产品的新规于7月15日生效，监管部门并未观望：在实施后头三周内即开出12张罚单，合计约420万元人民币。企业主要因两点被处罚——未清楚告知用户其互动对象是机器，以及未对与具有说服力的AI角色互动的用户进行年龄核验。如果你在中国提供此类产品，清晰标识与年龄核验如今已是强制要求，而且执法已经到来。
美国加密监管正处于成败的关键时刻
《CLARITY法案》——这部将最终理清美国由哪个监管机构（SEC或CFTC）监管哪类数字资产的法案——本周进入参议院全院表决阶段。它在8月休会前是通过还是搁置，将重塑整个美国加密行业格局。对Web3企业而言，明智之举不是观望，而是现在就把两种结果下的业务形态都推演清楚，做到有备无患。
此外本周（详见下文欧盟部分）：欧盟《AI法案》已于8月2日起可强制执行——若你面向欧盟市场，这一点很重要。
本周全球监管动态
AI监管观察
美国｜AI训练数据正成为法律主战场
对美国企业而言，最重要的AI议题仍是围绕“模型如何训练”的一批版权诉讼。针对OpenAI等公司的案件正处于证据开示阶段，另有一起判决已判定某AI法律检索工具因抄袭竞争对手内容而败诉。本周一家欧洲法院又以判决强化了这一趋势，判Suno音乐应用败诉（详见欧盟部分）。核心逻辑是：“我们用了网上能找到的一切来训练”并不是一个安全的答复。任何开发AI的美国企业，都应能够说明训练数据的来源、证明自己有权使用，并保留相关证据。
美国｜监管持续打击利用AI的诈骗
美国证券监管机构持续打击利用AI实施的诈骗——包括深度伪造、伪造语音和身份冒充——其中一起案件涉及从普通投资者处骗取超过1400万美元。对合规经营的企业而言，启示是：随着AI让骗局更加逼真，监管机构会要求你拿出更有力的证据，证明“你确实是你”。
美国（州层面）｜加州与纽约树立AI安全标准
加州SB 53已经生效，要求大型AI开发者公开安全方案并报告安全事件，每次违规最高罚款100万美元。纽约《RAISE法案》将于2027年1月1日生效，且刻意与加州思路保持一致——因此为其中一地做好合规，基本也能覆盖另一地。无论联邦层面的争论结果如何，这些州法都照常适用。
中国｜AI陪伴新规下的首批罚单（8月初）
中国针对拟人化与AI陪伴产品的新规于7月15日生效，监管部门很快展开执法——头三周即开出12张罚单，合计约420万元人民币。让企业“中招”的两点是：未清楚表明用户在与机器对话，以及未核验与具有说服力的AI角色互动的用户年龄。中国还将AI“智能体”（能采取行动、而不仅是聊天的系统）作为一个独立类别，并在医疗、交通、公共安全等敏感领域附加更多义务。如果你在中国提供这些产品，标识与年龄核验并非可选项。
欧盟｜欧洲AI规则动真格（8月2日）
欧盟《AI法案》成为法律已有一段时间，但8月2日才是其最重要的规则真正可被强制执行之日。自即日起，欧洲监管机构可对用于敏感领域的AI系统采取行动——招聘、信贷与保险决策、生物识别、关键基础设施、教育、执法等。这些“高风险”系统的提供者必须建立安全流程、技术文档与人工监督机制；仅仅使用这些系统的企业也负有义务。任何向欧盟用户展示AI生成或合成内容的企业，如今都必须清楚标注。此前有推迟部分规则的讨论，但推迟并未成为法律——因此应以2026年8月2日为准。
欧盟／德国｜法院裁定：AI训练并非“免费”（7月31日）
慕尼黑一家法院裁定，热门AI音乐应用Suno未经许可、使用德国音乐版权机构的歌曲进行训练，构成侵权，并须为其学习所用的歌曲以及生成的音乐两端付费获得授权。这是迄今最清晰的信号之一：未经授权的训练数据会带来真实的、跨境的法律风险，也强化了美国已在进行的多起版权之争。
Web3监管观察
美国｜加密“规则手册”的重大考验（8月3日）
《CLARITY法案》将在“商品”（由CFTC监管）与“证券”（由SEC监管）之间为各类数字资产划出清晰界线——这是行业期待多年的事。本周参议院领袖确认，该法案将在8月休会前进行全院表决，尽管要真正通过仍面临艰难的票数门槛。华尔街主要机构已表示支持。但目前尚未成为法律，因此宜将其视为提前准备的时机，而非庆祝的理由。
美国｜监管推动市场运行方式现代化
监管机构正朝着全天候交易和新型产品方向推进，相关公开征求意见期开放至8月26日。这些虽是渐进步骤，却显示出方向：美国市场正在被重塑，以适应加密资产“永不停歇”的特性。
美国（州层面）｜对预测市场与诈骗的反制
纽约州总检察长敦促国会加强对加密诈骗的保护，并指出相关投诉在三年内几乎增至三倍。与此同时，法院正在审理：联邦牌照是否允许预测市场平台绕开州级博彩法。启示是：联邦的“放行”并不总能让你免受各州的约束。
香港｜稳定币的高门槛
香港的稳定币发牌制度已经运作，但相当严格：在36家申请企业中，首轮仅两家（汇丰，以及一家由渣打支持的合资企业）获批。有意发行者需具备可观的资本，并以安全、流动性高的资产100%背书每一枚代币。如果香港在你的规划之中，请为一套要求严苛的申请流程做好准备。
欧盟｜没有牌照，就不能在欧盟开展加密业务（MiCA）
欧洲加密“规则手册”（MiCA）的过渡期已经结束：自7月1日起，面向欧盟客户的企业必须持有相应牌照，否则须停止经营。罚款最高可达年营业额的12.5%。如果欧盟是你的市场，请把牌照视为前提条件，而非事后补办的事项。
全球监管雷达｜本周一览
阴影行为本周新增，其余为持续跟踪的事项。
持续中｜美国（法院）
针对OpenAI等公司的AI版权与合理使用诉讼
当前状态：进行中
对企业意味着：这些案件将为“哪些训练数据被允许”定下基本规则。现在就保留好数据来源记录。
2026年8月3日｜美国
《CLARITY法案》（加密规则手册）进入参议院表决
当前状态：表决在即
对企业意味着：本周将决定美国由谁监管哪类加密资产。请为通过与搁置两种结果都做好准备。
2026年8月26日｜美国
CFTC全天候交易及永续合约方案的意见征询截止日
当前状态：截止日
对企业意味着：若全天候期货与你相关，请在窗口关闭前提交意见。
7月20–27日｜美国（纽约/华州）
各州对预测市场与加密诈骗的反制
当前状态：进行中
对企业意味着：联邦牌照未必能让你免受州级博彩或消费者保护法约束。请逐州评估风险敞口。
已生效｜美国（加州）
SB 53——前沿AI透明度法
当前状态：已生效
对企业意味着：大型AI开发者须公开安全方案并报告事件。每次违规最高100万美元。
2027年1月1日｜美国（纽约）
《RAISE法案》——纽约前沿AI安全法
当前状态：已通过
对企业意味着：与加州规则相似；现在就着手准备，一次搭建即可覆盖两地。
8月初｜中国
中国AI陪伴新规下的首批罚单（12张，420万元人民币）
当前状态：已开始执法
对企业意味着：若你在中国运营AI陪伴或拟人化产品，请清晰标识并核验用户年龄——监管已在处罚这些疏漏。
2026年7月15日｜中国
AI陪伴与AI智能体规则生效；内容标识标准
当前状态：已生效
对企业意味着：AI须表明自己非真人、保护未成年人，并对AI生成内容加注显式+隐式标识。
2025年 / 2026年4月｜香港
稳定币发牌落地；首两家发行人获批（汇丰、Anchorpoint）
当前状态：已生效
对企业意味着：门槛很高——36家仅2家通过。申请前请确认资本与100%储备要求。
2026年8月2日｜欧盟
欧洲《AI法案》规则正式启用并可强制执行
当前状态：现已生效
对企业意味着：若你的AI涉及招聘、信贷、保险或生物识别，或只是面向欧盟——都须标注AI内容并备齐文档。监管现在可以处罚你。
2026年7月31日｜欧盟／德国
慕尼黑法院裁定Suno因训练使用歌曲构成侵权
当前状态：已判决
对企业意味着：使用未授权内容训练AI可能付出代价——训练与生成两端都要付。
2026年7月1日｜欧盟
MiCA过渡期结束, 罚款开始
当前状态：已生效
对企业意味着：没有牌照，就没有欧盟加密业务。罚款最高可达年营业额的12.5%。
范川理律师观察
如果说本周的种种变化有一条共同主线，那就是：科技已经进入“规则时代”。
过去几年，无论AI还是Web3，竞争大多围绕原始能力展开——更大的模型、更快的产品、更多的用户。目标是抢先起跑，其余的以后再说。这个时代正在落幕。随着世界各国相继完成各自的规则体系，企业面临的问题正从“我们能不能做出来”，转向“我们能否在每一个所处市场里，合法且长期地把它运营下去”。
在AI领域，监管的关注点正从“模型能生成什么”，扩展到数据从何而来、如何治理、如何保护用户，以及当AI自主行动时会发生什么。AI陪伴、AI智能体等新型产品带来一个更棘手的问题：不只是AI“说了什么”，而是它“做了什么”——以及当它这么做时，责任在谁。
在Web3领域，主要市场——美国、香港、欧盟——都在从“个案执法”转向清晰、成文的框架。今后的赢家，不会是那些去寻找监管最宽松地区的企业，而是那些能在真实规则之下、跨越国界稳健运营的企业。
对于出海拓展的企业而言，这是一个新阶段。过去，先讲市场机会、再把法律细节留到以后处理就足够了；如今，法律架构、监管路径与风险管理，本身就是产品设计的一部分。真正具备全球竞争力的企业，不只是能快速推出产品，更能理解每个市场的规则，并在规则框架内依然找到创新的空间。
——范川理律师
本周企业合规行动建议
如果你是AI企业
把你的产品分类。弄清楚哪些AI产品属于陪伴型或拟人化型，哪些涉及高风险决策（招聘、信贷、医疗、教育），哪些面向儿童或欧盟用户。你的义务取决于你落在哪个类别。
把数据“家底”理清楚。对每一套数据集，都要能说明其来源、证明你有权使用——并保留证据。在Suno判决之后，这是你最重要的一道保护。
现在就为欧洲做好准备。若你面向欧盟销售，今天就把内容标识、技术文档和人工监督环节落实到位——规则已经生效，而非“即将到来”。
如果你在中国经营， 请确保陪伴型与拟人化AI清楚表明自己是机器、加入安全提示，并核验用户年龄。监管已在处罚跳过这些步骤的企业。
如果你是Web3企业
关注美国表决，并为两种结果做准备。《CLARITY法案》本周通过还是搁置，都会改变你的监管处境。请把你的代币与业务在两种情形下的样子都勾画出来。
搞清楚你的代币“是什么”。将每种代币归类为商品、稳定币、收藏品、工具或证券，并能说明理由。这一判断决定你要向哪个监管机构负责。
别以为联邦牌照能让你免受各州约束。若你经营预测市场或事件合约，请逐州评估博彩与消费者保护方面的风险敞口。
如果香港在你的版图上， 请在申请前确认自己能满足资本与100%储备要求——获批门槛很高。
收紧反欺诈与身份核验措施。在诈骗上升、监管紧盯的当下，健全的反洗钱、了解你的客户（KYC）与信息披露实践，如今已是基本门槛。
下周关注
AI： 规则生效后欧盟《AI法案》执法的进展，以及美国AI版权案件的下一步动向。
Web3： 《CLARITY法案》能否在8月休会前通过参议院——这是今年美国加密领域最大的变量。
本简报以通俗语言撰写，仅供一般了解之用，不构成法律意见。如涉及AI治理或Web3合规的具体问题，请咨询我们专业律师。
AI & Web3 Global Regulatory Weekly
Cross-Border Compliance Watch for Tech Companies Going Global
Week in Review: July 30 – August 5, 2026
Three Things That Matter This Week
For U.S. AI companies: where your training data comes from is now a core business risk — not a technical detail.
U.S.courts are deep into a wave of copyright cases over how AI models are trained, and this week a court in Europe drove the point home by ruling that the music app Suno must pay to license the songs it trained on — for both the training and the output. The message for any American company building AI is the same: if you can't prove your training data was obtained legally, that's a business risk that can cost real money. Companies that keep clean records of where their data came from will be far better protected than those that don't.
In China, companion and human-like AI is now being fined — and enforcement is moving fast.
China's new rules for human-like and companion AI took effect on July 15, and regulators didn't wait: they issued 12 fines totaling about 4.2 million RMB in the first three weeks.Companies got caught for two things — not making it obvious that users were talking to a machine, and not checking the ages of users interacting with persuasive AI characters. If you offer this kind of product in China, clear labeling and age checks are now mandatory, and enforcement is already here.
U.S. Crypto rules are at a make-or-break moment.
The CLARITY Act — the bill that would finally settle which U.S.regulator (the SEC or the CFTC) oversees which digital assets — reached the Senate floor this week. Whether it passes or stalls before the August break will reshape the entire U.S. Crypto landscape. For Web3 companies, the smart move isn't to wait and see. It's to map out how your business looks under both outcomes, so you're ready either way.
Also this week (see the EU section below): Europe's AI Act became enforceable on August 2 — important if you sell into the EU.
This Week's Global Developments
AI Watch
USA — AI Training Data Is Now a Legal Battleground
The biggest AI story for U.S.companies remains the wave of copyright lawsuits over how models are trained. Cases against OpenAI and others are moving through discovery, and a separate ruling has already gone against an AI legal-research tool for copying a competitor's content. This week a European court reinforced the trend by ruling against the music app Suno (see the EU section). The through-line: “we trained on whatever was available online” is not a safe answer. Any U.S. company building AI should be able to show where its training data came from and that it had the right to use it — and should keep the proof.
USA — Regulators Keep Going After AI-Powered Scams
U.S. securities regulators are continuing to crack down on fraud that uses AI — deepfakes, fake voices and impersonation — including a case involving more than $14 million taken from everyday investors. The takeaway for legitimate businesses: as AI makes scams more convincing, expect regulators to demand stronger proof that you are who you say you are.
USA (States) — California and New York Set the AI Safety Bar
California's SB 53 is already in force, requiring large AI developers to publish safety plans and report incidents, with penalties up to $1 million per violation. New York's RAISE Act follows on January 1, 2027 and is deliberately aligned with California's approach — so if you build for one, you can build for both. These state rules apply regardless of where the federal debate lands.
China — First Fines Under the New AI-Companion Rules (early August)
China's new rules for human-like and companion AI took effect on July 15, and regulators didn't wait long to enforce them — issuing 12 fines totaling about 4.2 million RMB in the first three weeks. The two things that got companies in trouble: not making it obvious that users were talking to a machine, and not checking the ages of users interacting with persuasive AI characters. China also treats AI “agents” (systems that take actions, not just chat) as a distinct category with extra duties in sensitive fields like health, transport and public safety. If you offer these products in China, labeling and age checks are not optional.
EU — Europe's AI Rules Get Real (August 2)
The EU's AI Act has been law for a while, but August 2 is the day its most important rules actually became enforceable. From now on, European regulators can act against AI systems used in sensitive areas — hiring, credit and insurance decisions, biometric identification, critical infrastructure, education, policing and more. Providers of these “high-risk” systems must have safety processes, documentation and human oversight in place; companies that simply use them have duties too. And any business showing AI-generated or synthetic content to EU users must now clearly label it. There's been talk of delaying some of this, but that delay has not become law — so August 2, 2026 is the date to plan around.
EU / Germany — A Court Says AI Training Isn't Free (July 31)
A court in Munich ruled that Suno, a popular AI music app, broke copyright law by training on songs from Germany's music-rights body without permission — and that it has to pay to license both the songs it learned from and the music it produces. This is one of the clearest signals yet that unlicensed training data carries real, cross-border legal exposure, and it reinforces the copyright fights already underway in the U.S.
Web3 Watch
USA — The Crypto Rulebook's Big Test (August 3)
The CLARITY Act would draw a clear line between which digital assets are treated as “commodities” (overseen by the CFTC) and which are “securities” (overseen by the SEC) — something the industry has wanted for years.This week the Senate leader confirmed it will get a floor vote before the August break, though it still faces a tough numbers hurdle to actually pass. Major Wall Street firms have backed it. Nothing is law yet, so treat this as a moment to prepare, not to celebrate.
USA — Regulators Modernize How Markets Run
Regulators are moving toward round-the-clock trading and new product types, with a public comment window open until August 26. These are incremental steps, but they show the direction of travel: U.S. markets are being reshaped to accommodate the always-on nature of crypto.
USA (States) — Pushback on Prediction Markets and Scams
New York's attorney general urged Congress to toughen crypto-scam protections, noting fraud complaints have nearly tripled in three years. Meanwhile, courts are testing whether a federal license lets prediction-market platforms sidestep state gambling laws. The lesson: a federal green light doesn't always protect you from individual states.
Hong Kong — A High Bar for Stablecoins
Hong Kong's stablecoin licensing system is up and running, but it's selective: of 36 companies that applied, only two (HSBC and a Standard Chartered-backed venture) were approved in the first round. Would-be issuers need substantial capital and must back every coin with safe, liquid reserves. If Hong Kong is on your roadmap, budget for a demanding application process.
EU — No License, No EU Crypto Business (MiCA)
Europe's crypto rulebook (MiCA) has finished its transition period: as of July 1, companies serving EU customers must hold a proper license or stop. Penalties can reach up to 12.5% of annual revenue. If the EU is a market for you, treat licensing as a prerequisite, not an afterthought.
Global Regulatory Radar — This Week at a Glance
Shaded rows are new this week. The rest are ongoing items worth tracking.
Ongoing｜USA (Courts)
AI copyright/fair-use suits vs. OpenAI and others
Status：Active
Business impact：These cases will set the ground rules for what training data is allowed. Keep clean records of where your data came from now.
Aug 3, 2026｜USA
The CLARITY Act (crypto rulebook) heads to a Senate vote
Status：Vote imminent
Business impact：This week decides who regulates which crypto assets in the U.S. Plan for both passage and failure.
Aug 26, 2026｜USA
Deadline to comment on CFTC's 24/7 trading & perpetuals plan
Status：Deadline
Business impact：If round-the-clock futures matter to you, speak up before the window closes.
Jul 20–27｜USA (NY/WA)
States push back on prediction markets & crypto scams
Status：Active
Business impact：A federal license may not shield you from state gambling or consumer-protection law. Check state-by-state exposure.
In force｜USA (CA)
SB 53 — frontier-AI transparency law
Status：In force
Business impact：Big AI developers must publish safety plans and report incidents. Up to $1M per violation.
Jan 1, 2027｜USA (NY)
RAISE Act — New York's frontier-AI safety law
Status：Enacted
Business impact：Similar to California's rules; start preparing now so you build once for both.
Early Aug｜China
First fines under China's AI-companion rules (12 fines, RMB 4.2M)
Status：Enforcement started
Business impact：If you run companion or human-like AI in China, label it clearly and check users' ages — regulators are already fining these gaps.
Jul 15, 2026｜China
AI-companion & AI-agent rules in force; content-labeling standard
Status：In force
Business impact：AI must identify itself as non-human, protect minors, and carry visible + hidden labels on AI-made content.
2025 / Apr 2026｜Hong Kong
Stablecoin licensing live; first two issuers approved (HSBC, Anchorpoint)
Status：In force
Business impact：The bar is high — only 2 of 36 applicants passed. Confirm capital and 100% reserve rules before applying.
Aug 2, 2026｜EU
Europe's AI Act rules are now switched on and enforceable
Status：Now in force
Business impact：If your AI touches hiring, lending, insurance or biometrics — or shows up in the EU at all — you must label AI content and have your paperwork ready. Regulators can now fine you.
Jul 31, 2026｜EU / Germany
A Munich court rules Suno broke copyright law by training on songs
Status：Ruling issued
Business impact：Using content you didn't license to train AI can cost you — for both the training and the output.
Jul 1, 2026｜EU
MiCA transition window closed and fines are imposing.
Status：In force
Business impact：No license, no EU crypto business. Fines reach up to 12.5% of yearly revenue.
Joanna Fan's Take
If there's one thread connecting everything this week, it's this: technology has entered the age of rules.
For the past few years, both AI and Web3 competed mostly on raw capability — bigger models, faster products, more users. The goal was to move first and figure out the rest later. That era is closing. As governments around the world finish building their rulebooks, the question companies face is changing from “can we build it?” to “can we run it, legally and for the long term, in every market we're in?”
In AI, oversight is widening from what a model can generate to where its data came from, how it's governed, how it protects users, and what happens when it acts on its own. New products like AI companions and AI agents force a harder question: not just what the AI says, but what it does — and who's responsible when it does it.
In Web3, the major markets — the U.S., Hong Kong, the EU — are all moving from case-by-case enforcement toward clear, written frameworks. Going forward, the winners won't be the ones who find the loosest jurisdiction. They'll be the ones who design a business that can operate cleanly under real rules, across borders.
For companies expanding abroad, this is a new phase. It used to be enough to lead with the market opportunity and sort out the legal details later. Now, legal structure, regulatory path and risk management are part of the product design itself. The companies that will win globally aren't just the ones that ship fast — they're the ones that understand each market's rules and still find room to innovate inside them.
— Joanna Fan, your legal observer
What to Do This Week
If You're an AI Company
Sort your products into buckets.Figure out which of your AI products are companions or human-like, which make high-stakes decisions (hiring, credit, health, education), and which reach children or European users. Your obligations depend on which bucket you're in.
Get your data house in order.Be able to show, for every dataset, where it came from and that you had the right to use it — and keep the proof. After the Suno ruling, this is your single biggest protection.
Prepare for Europe now.If you sell into the EU, put content labels, documentation and human-oversight steps in place today — the rules are live, not coming.
If you operate in China, make sure companion and human-like AI clearly identifies itself as a machine, adds safety reminders, and verifies users' ages.Regulators are already fining companies that skip these steps.
If You're a Web3 Company
Watch the U.S.vote and plan for both outcomes. Whether the CLARITY Act passes or stalls this week changes your regulatory picture. Sketch out how your tokens and business look under each scenario.
Know what your token is.Classify each token as a commodity, stablecoin, collectible, tool or security, and be ready to explain why. That analysis drives which regulator you answer to.
Don't assume a federal license protects you from states.If you run prediction markets or event contracts, check gambling and consumer-protection exposure state by state.
If Hong Kong is on your map, make sure you can meet the capital and 100%-reserve requirements before you apply — the approval bar is high.
Tighten fraud and identity controls.With scams rising and regulators watching, strong anti-money-laundering, know-your-customer and disclosure practices are now table stakes.
Next Week to Watch
AI: The outcome of any EU AI Act enforcement moves now that the rules are live, and the next steps in the U.S.AI copyright cases.
Web3: Whether the CLARITY Act clears the Senate before the August recess — the single biggest swing factor for U.S.crypto this year.
This briefing is written in plain language for general understanding and does not constitute legal advice. For AI governance or Web3 compliance questions, please consult with our qualified lawyers.
/ END.
*本文为上海曼昆律师事务所的原创文章，仅代表本文作者个人观点，不构成对特定事项的法律咨询和法律意见。如需转载及法律咨询，请添加客服：mankunlawyer
