AI Regulation & Policy — 2026年8月24日 週次レポート
AI Regulation & Policyのニュース＆アップデート — すべての記述に一次ソースのリンク付き。
重要な発見
エグゼクティブサマリー（5件）
- •The EU AI Act has fully transitioned from legislative preparation to active enforcement: transparency rules are being enforced as of August 2, Prohibition 9 has a confirmed December 2026 deadline, and the AI Omnibus has entered into force. Organizations that have not completed transparency compliance are now in violation, not merely behind schedule. The enforcement calendar is accelerating, not slowing.
- •A children's online safety enforcement wave is cresting simultaneously across multiple jurisdictions: New Jersey signed the Kids Code Act, New York finalized SAFE for Kids Act rules effective January 2027, Hawaii enacted an AI Disclosure and Safety Act, and TikTok agreed to a $400 million COPPA settlement. The pattern — legislation, then enforcement, then massive settlements — is now a documented cycle that AI-powered consumer platforms must treat as a near-term operational risk.
- •The White House's authorization of private-sector offensive cyber operations and the French CNIL's agentic AI guidance represent two ends of the same governance challenge: as AI systems become capable of autonomous action, governments are simultaneously expanding the authorized scope of AI-enabled operations and scrambling to define the legal frameworks that constrain them. The gap between AI capability and governance infrastructure is widening, not closing.
- •The legal AI market is undergoing a structural shift from tool adoption to platform depth: Harvey's launch of a proprietary legal AI model and memory-enabled platform, Twin1's digital twin approach to knowledge capture, and Avvoka's template automation partnership with Harvey all signal that the competitive frontier has moved from 'does it work' to 'does it know your firm.' Organizations evaluating legal AI tools must now assess institutional knowledge integration, not just task performance.
- •Singapore is emerging as the definitive Asia-Pacific AI governance hub: AI fluency programmes for legal professionals, 200 MW of new data centre capacity allocated, and an IMDA-IAPP professional training partnership create a self-reinforcing ecosystem of governance talent, infrastructure, and regulatory clarity that will attract AI governance operations from across the region.
今回の要点（15件）
- 1.EU AI Act transparency rules are in active enforcement as of August 2, 2026, and Prohibition 9 covering non-consensual sexually explicit AI content takes effect December 2026 as part of the AI Omnibus. [6]
- 2.New Jersey signed the Kids Code Act on August 11, 2026, imposing privacy-by-default and safety-by-design obligations on platforms likely accessed by minors, joining New York's SAFE for Kids Act (effective January 25, 2027) and Hawaii's AI Disclosure and Safety Act in a state-level children's safety law wave. [4]
- 3.President Trump issued a National Security Presidential Memorandum on August 12, 2026 authorizing vetted private-sector companies to conduct offensive cyber operations against foreign cyber-enabled criminal organizations under federal oversight, with implementation guidance expected by mid-October 2026. [4]
- 4.The French CNIL published a joint exploratory note with the French AI and Digital Council on agentic AI and GDPR implications on July 20, 2026, joining the UK ICO, Spanish AEPD, and Singaporean IMDA in a rapidly accumulating body of regulator commentary on autonomous AI systems. [2]
- 5.Harvey introduced Tenet, its first post-trained open-weight AI model for legal work, and launched Harvey II with Memory capabilities, signaling a transition toward a vertically integrated legal AI stack. [10a]
- 6.Twin1 AI launched out of stealth with a $20 million seed round to build digital twins of law firm professionals, with strategic investment from Orrick and backing from Bessemer Venture Partners. [10b]
- 7.UK MHRA published guidance on July 29, 2026 clarifying that AI ambient scribing tools require a 'medical purpose' to qualify as medical devices, narrowing the compliance burden compared to prior NHS England interpretation. [2]
- 8.Singapore's IMDA launched AI fluency programmes for legal professionals on August 20, 2026 — AIxLegal by SAL and AIxLegal by SCCA — alongside nearly 2,000 tech job opportunities, deepening Singapore's position as Asia-Pacific's AI governance hub. [12]
- 9.Wellington Management published analysis at the Harvard Law School Forum on August 21, 2026 finding nearly three in four companies plan to deploy agentic AI within two years despite only one in five having a mature governance model for autonomous agents. [9a]
- 10.Tech Policy Press published multiple pieces on AI and the 2026 US midterm elections, including analysis of how AI is reshaping election information and a podcast on synthetic voters, signaling AI-generated content is entering the active election cycle. [3]
- 11.A lawsuit filed in the Northern District of California alleges AI-based tools used to score and select employees for a reduction in force discriminated against workers on protected leave and those with disabilities, marking an emerging litigation category. [5]
- 12.The EDPB adopted guidelines on web scraping in the context of generative AI alongside its anonymisation guidelines, with the anonymisation consultation open until October 30, 2026. [4]
- 13.The French Constitutional Council struck down France's under-15 social media ban on August 14, 2026, finding it disproportionate to the objective pursued and insufficiently protective of privacy rights in the age verification process. [2]
- 14.CSET published analysis on August 20, 2026 examining how the U.S. military's growing use of AI could affect military decision-making and human judgment, with a senior fellow warning that avoiding AI-driven dystopias requires maintaining control of human behavior as well as AI. [13a]
- 15.TikTok agreed to pay $400 million to resolve DOJ claims of illegally collecting personal information from children under 13, as reported by Law360 on August 22, 2026, marking one of the largest COPPA recoveries ever obtained. [14]
市場動向
AI Governance Moves from Compliance Project to Enterprise Foundation
This week's activity confirms AI governance is institutionalizing as a permanent enterprise function rather than a one-time compliance exercise. Wellington Management published analysis at the Harvard Law School Forum on August 21, 2026 noting that nearly three in four companies plan to deploy agentic AI within two years despite only one in five having a mature governance model for autonomous agents, framing this gap as a core business risk [9a]. Tech Policy Press published perspectives on Augus…
Legal AI Market Accelerates Toward Specialized Models and Platform Depth
The legal AI competitive landscape shifted materially this week with Harvey introducing Tenet, its first post-trained open-weight AI model specifically for legal work, trained through asynchronous reinforcement learning on synthetic, publicly available legal, and human expert data [10a]. Twin1 AI launched out of stealth with a $20 million seed round to build digital twins of law firm professionals, with investment from Orrick and backing from Bessemer Venture Partners and Tribeca Venture Partner…
AI and Elections: Synthetic Content Risk Enters Midterm Cycle
Tech Policy Press published multiple pieces this week on AI's intersection with the 2026 U.S. midterm elections, including a podcast on August 16 with the Brennan Center's Larry Norden on how AI is reshaping election information, and a newsletter on August 16 titled 'AI Meets the US Midterm Elections' [3]. A separate podcast titled 'Here Come the Synthetic Voters' was published August 16 [3]. Tech Policy Press also published analysis on August 17 examining whether China will crack down on open-w…
競合動向
Harvey Deepens Legal AI Moat with Proprietary Model and Platform Expansion
Harvey made two significant competitive moves this week. First, it introduced Tenet, its first post-trained open-weight AI model for legal, which Harvey's research showed performing competitively against frontier models from Anthropic, OpenAI, and Google across legal agent and knowledge benchmarks [10a] (company announcement — may reflect promotional framing). Second, Harvey launched Harvey II, the next generation of its platform with Memory at its core, supporting user preferences, matter histo…
Singapore Positions as Asia-Pacific AI Governance Hub with Legal Sector Focus
Singapore's IMDA launched two AI fluency programmes for legal professionals on August 20, 2026 — AIxLegal by the Singapore Academy of Law and AIxLegal by the Singapore Corporate Counsel Association — alongside nearly 2,000 job opportunities for Singapore's tech workforce [12]. IMDA also provisionally allocated 200 megawatts of new data centre capacity to four operators through its second Data Centre Call for Application on August 21, 2026 [12]. These moves, combined with the earlier IMDA-IAPP AI…
AI in Workforce Decisions Generates First Discrimination Litigation Wave
National Law Review reported on August 19, 2026 that a lawsuit filed in the Northern District of California alleges AI-based tools used to score, rank, and select employees for a reduction in force discriminated against workers on protected leave and those with disabilities [5]. This follows the National Law Review's August 18 analysis of how AI-driven productivity monitoring tools can trigger New York City's Automated Employment Decision Tool law and other privacy requirements [5a]. The pattern…
制度・規制動向
EU AI Act Enforcement Deepens: Transparency Rules Now Active, Prohibition 9 Approaching
The EU AI Act's enforcement trajectory continued to deepen this week. The EU Digital Strategy confirmed the Commission began enforcing AI Act rules and new transparency requirements on August 2, 2026, and the AI Act page confirmed transparency rules came into effect in August 2026 [7]. The EU AI Act page also confirmed that Prohibition 9 — covering AI systems that generate non-consensual sexually explicit content — comes into effect in December 2026 as part of the AI Omnibus [6]. The Hunton Priv…
French CNIL Publishes Agentic AI and GDPR Note — Regulator Commentary Accumulates Globally
On July 20, 2026, the French data protection authority CNIL published a joint exploratory note with the French AI and Digital Council on the data protection implications of agentic AI, identifying where GDPR may come under strain when processing personal data via autonomous systems and suggesting legal and technical mitigation measures. Global Policy Watch reported on this development on August 20, 2026, noting the CNIL framed the exercise as connected to G7 data protection authority engagement …
US State Children's Online Safety Laws Proliferate — New Jersey Kids Code Act Signed
New Jersey Governor Mikie Sherrill signed the Kids Code Act on August 11, 2026, imposing privacy-by-default and safety-by-design obligations on online service providers whose services are reasonably likely to be accessed by minors, as reported by the Hunton Privacy Blog on August 20, 2026 [4]. The New York Attorney General released final rules implementing the Stop Addictive Feeds Exploitation for Kids Act on July 28, 2026, which takes effect on January 25, 2027, as reported by both the Hunton P…
UK MHRA Clarifies AI Medical Device Scope — Ambient Scribing Tools Narrowed
On July 29, 2026, the UK's Medicines and Healthcare products Regulatory Agency published new guidance on ambient voice technology-enabled products, clarifying that summarization alone is not sufficient to trigger medical device requirements and confirming that ambient scribing tools must have a 'medical purpose' to fall within the medical device definition. Global Policy Watch reported this guidance was intended to overrule a prior NHS England decision that had classified AI summarization tools …
White House Authorizes Private-Sector Offensive Cyber Operations Against Criminal Organizations
On August 12, 2026, President Trump issued a National Security Presidential Memorandum establishing a federally supervised program enabling vetted private-sector companies to conduct offensive cyber operations against foreign cyber-enabled transnational criminal organizations, under U.S. government control and oversight. The Hunton Privacy Blog reported on August 20, 2026 that additional implementation guidance is expected by mid-October 2026 [4]. Global Policy Watch reported on August 21, 2026 …
EDPB Anonymisation Guidelines Consultation Ongoing — AI Training Data Implications Persist
The EDPB's draft Guidelines 02/2026 on Anonymisation, adopted July 7, 2026 and open for public consultation until October 30, 2026, remain an active regulatory development with direct implications for AI training data practices under GDPR. The Hunton Privacy Blog reported on August 17, 2026 that the EDPB also adopted guidelines on web scraping in the context of generative AI and finalized guidelines on blockchain data processing [4]. Global Policy Watch reported on August 18, 2026 that the new a…
Singapore Launches AI Fluency for Legal Professionals — Governance Workforce Capacity Building Accelerates
Singapore's IMDA, together with the Singapore Academy of Law and the Singapore Corporate Counsel Association, launched two AI fluency programmes for legal professionals on August 20, 2026 — AIxLegal by SAL and AIxLegal by SCCA — alongside nearly 2,000 job opportunities for Singapore's tech workforce [12]. This follows the IMDA-IAPP Memorandum of Intent signed July 22, 2026 to expand AI governance professional training. The IAPP reported on August 20, 2026 that the Australian OAIC's actions signa…
ソース活動
先週からの変化
EU AI Act Transparency Enforcement Active — Prohibition 9 Deadline Confirmed December 2026
EU AI Act transparency rules are now in active enforcement as of August 2, 2026, and the EU AI Act page confirmed Prohibition 9 (non-consensual sexually explicit AI content) takes effect December 2026 as part of the AI Omnibus. The compliance window has closed; enforcement is underway. [6] [7]
New Jersey Kids Code Act Signed — State Children's AI Safety Laws Multiply
New Jersey Governor signed the Kids Code Act on August 11, 2026, imposing privacy-by-default and safety-by-design obligations on platforms likely accessed by minors. Combined with New York's SAFE for Kids Act final rules (effective January 25, 2027) and Hawaii's AI Disclosure and Safety Act (July 14, 2026), a wave of state children's online safety laws is creating layered compliance obligations for AI-powered consumer platforms. [4]
Harvey Launches Tenet — First Legal-Specific Post-Trained Open-Weight AI Model
Harvey introduced Tenet on August 20, 2026, its first post-trained open-weight AI model for legal work, trained on synthetic, publicly available legal, and human expert data. Harvey also launched Harvey II with Memory capabilities. These moves signal Harvey is building a vertically integrated legal AI stack, raising competitive pressure across the legal AI market. [10a]
White House NSPM Authorizes Private-Sector Offensive Cyber Operations
President Trump issued a National Security Presidential Memorandum on August 12, 2026 establishing a federally supervised program for vetted private-sector companies to conduct offensive cyber operations against foreign cyber-enabled criminal organizations. Implementation guidance expected by mid-October 2026. This creates a new regulatory category for AI-enabled cybersecurity companies. [4] [2]
ウォッチリスト — 今後の締切
EU AI Act Prohibition 9 (non-consensual sexually explicit AI content) takes effect
ソース: EU AI ActNew York SAFE for Kids Act (Stop Addictive Feeds Exploitation) takes effect
ソース: Hunton Privacy Blog示唆・見るべき論点（11件）
- 1.The confirmation of Prohibition 9's December 2026 effective date means AI companies have approximately four months to audit their systems for non-consensual intimate image generation capabilities — a compliance deadline that is closer than most organizations' annual planning cycles. [6]
- 2.The French CNIL's agentic AI note, joined by the UK ICO, Spanish AEPD, and Singaporean IMDA, represents a coordinated global regulator signal: agentic AI governance guidance is accumulating faster than binding rules, and organizations that wait for binding requirements before acting will face a compressed compliance window. [2]
- 3.The White House NSPM authorizing private-sector offensive cyber operations creates a novel compliance category: AI-enabled cybersecurity companies that seek to participate in the program will need to navigate federal vetting, authorization, and oversight requirements that do not yet exist in final form — implementation guidance expected by mid-October 2026 will be critical. [4]
- 4.Harvey's launch of Tenet as an open-weight model is strategically significant: by releasing a legal-specific open-weight model, Harvey enables law firms to build their own models on top of Tenet, creating a platform ecosystem dynamic that could lock in enterprise customers more durably than a closed-model approach. [10a]
- 5.Wellington Management's finding that 74% of companies plan agentic AI deployment within two years but only 20% have mature governance models for autonomous agents quantifies the governance gap that regulators like the CNIL are already addressing — organizations in the 80% without mature governance are on a collision course with forthcoming binding requirements. [9a]
- 6.The TikTok $400 million COPPA settlement, combined with New Jersey's Kids Code Act and New York's SAFE for Kids Act, signals that children's privacy enforcement is entering a phase of large-scale financial consequences — not just injunctive relief. AI platforms with minor users should treat the TikTok settlement as a benchmark for their own potential exposure. [14]
- 7.The UK MHRA's clarification that ambient scribing tools require a 'medical purpose' to qualify as medical devices resolves a significant compliance uncertainty for AI healthcare companies operating in Great Britain — but the underlying question of where AI clinical documentation tools sit on the medical device spectrum remains live in other jurisdictions. [2]
- 8.The French Constitutional Council's invalidation of the under-15 social media ban on proportionality and privacy grounds establishes a judicial precedent that age verification requirements must themselves be privacy-protective — a constraint that will complicate implementation of similar laws in other EU member states. [2]
- 9.Singapore's simultaneous launch of AI fluency programmes for legal professionals and allocation of 200 MW of new data centre capacity reflects a deliberate national strategy to capture both the governance talent and infrastructure dimensions of the AI economy — a model other Asia-Pacific jurisdictions will likely emulate. [12]
- 10.The emerging litigation pattern of AI workforce decision tools generating discrimination claims — as seen in the Northern District of California RIF case — means organizations using AI for employment decisions must now treat bias audit documentation as litigation preparation, not just compliance evidence. [5]
- 11.CSET's analysis of how AI could hollow out U.S. military decision-making by eroding human judgment, published in Foreign Affairs on August 20, 2026, signals that the national security AI governance debate is shifting from 'how do we deploy AI faster' to 'how do we preserve human control as AI becomes more capable' — a reframing that will influence both military procurement and civilian AI governance standards. [13a]
信頼度サマリー
今週引用したソース 19 件あなたが選んだ 30 件の監視URLから検出（1つのURLから複数記事が出ることがあります）。
各ソースは信頼度レベルに応じて重み付けされています。単独ソースの主張は AI 合成時に未検証としてフラグ付けされます。
参照ソース一覧
Source for Commission beginning enforcement of AI Act transparency rules on August 2, 2026; AI Gigafactories call with more than €30 billion in investment; approximately 190 organisations signing Code of Practice on Transparency of AI-generated Content; AI Omnibus entry into force July 27; Digital Europe Programme data projects update August 20.
Source for French CNIL agentic AI and GDPR note (July 20, 2026); UK MHRA ambient scribing guidance (July 29, 2026); French Constitutional Council striking down under-15 social media ban (August 14, 2026); White House NSPM on private-sector offensive cyber operations (August 12, 2026); EDPB anonymisation guidelines consultation open until October 30, 2026; New York SAFE for Kids Act final rules (July 28, 2026).
Source for AI and 2026 US midterm elections analysis (August 16); synthetic voters podcast (August 16); China open-weight model crackdown analysis (August 17); AI audits need a power test perspective (August 20); Congress AI whistleblower protection perspective (August 20); Vietnam cyber law enforcement (August 17).
Source for New Jersey Kids Code Act signed August 11, 2026; New York SAFE for Kids Act final rules effective January 25, 2027; Hawaii AI Disclosure and Safety Act signed July 14, 2026; White House NSPM on private-sector cyber operations (August 12, 2026); CalPrivacy settlements with LocateSmarter and Cybba (August 11-13, 2026); EDPB web scraping and anonymisation guidelines; FTC suit against Hims & Hers.
Source for AI workforce discrimination lawsuit in Northern District of California (August 19, 2026); AI-powered employee monitoring and NYC AEDT law analysis (August 18, 2026); government contractor AI compliance guide; Pennsylvania data centre executive order tying compliance to GRID requirements (August 21, 2026); DOL proposed safe harbor for electronic group health plan disclosures.
Source for EU AI Act risk-based framework; transparency rules effective August 2026; Prohibition 9 effective December 2026 as part of AI Omnibus; high-risk AI system obligations effective December 2027; GPAI rules effective August 2025.
Corroborating source for EU AI Act enforcement commencement August 2, 2026 and AI Omnibus entry into force July 27, 2026.
Source for EU proposed CADA cloud sovereignty analysis (August 20); EU AML regulation personal data processing analysis (August 20); FTC enforcement policy on personalized pricing tools (August 20); OAIC accountability and governance focus in Asia-Pacific (August 20); EU data sovereignty plan (August 20); AI governance article count growth from 188 to 194 items during reporting period.
Source for Wellington Management AI governance for private companies analysis (August 21, 2026) finding 74% of companies plan agentic AI deployment within two years but only 20% have mature governance models; 2026 proxy season review; DOJ withdrawal of antitrust guidance for proxy advisory industry; SEC Spring 2026 regulatory agenda analysis.
Source for Harvey introducing Tenet post-trained open-weight AI model (August 20, 2026); Twin1 AI $20 million seed round (August 20, 2026); Avvoka Curate template tool launch (August 20, 2026); Texas court shielding generative AI review prompts as work product (August 20, 2026); pre-ILTACON legaltech rundown including NetDocuments and DISCO agentic e-discovery tool (August 21, 2026); Harvey II launch (August 18, 2026); prompt injections as court risk (August 18, 2026).
Source for Harvey II launch with Memory capabilities (August 18, 2026); Harvey-DeepL translation integration; Avvoka partnership with Harvey and Curate launch (August 20, 2026); Twin1 $20 million seed round (August 20, 2026); NetDocuments M&A due diligence expansion (August 20, 2026); DeepJudge-Legora bidirectional partnership (August 18, 2026).
Source for IMDA launching AI fluency programmes for legal professionals AIxLegal by SAL and AIxLegal by SCCA (August 20, 2026); IMDA and EDB provisionally allocating 200 MW of new data centre capacity to four operators through DC-CFA2 (August 21, 2026); MDDI Family 72 scholarships for digital leaders (August 12, 2026).
Source for CSET analysis on how AI could hollow out U.S. military decision-making (Foreign Affairs, August 20, 2026); Trump crypto firm backing venture offering AI from restricted Chinese companies (Reuters, August 17, 2026); CSET expert commentary on Trump AI strategy and China competitiveness.
Source for TikTok agreeing to pay $400 million to resolve DOJ COPPA claims for illegally collecting personal information from children under 13 (August 22, 2026); Meta whistleblower testimony in California trial on kids' safety (August 20, 2026).
Source for AISI incident report on unsanctioned agent behaviour during cyber testing (August 4, 2026); AISI blog updates on cyber evaluations and frontier AI trends.
Source for state consumer privacy law count update — 2026 started with 20 SCPLs, four new ones added in first half of 2026 (August 17, 2026); China simplified personal information protection regime for small-scale handlers (August 6, 2026).
Source for UN using AI to advance human rights (August 22, 2026); youth advocates gathering in New York to launch new AI standards (August 17-18, 2026); AI helping deliver food to hungry families in Somalia (featured story).
Source for UNESCO pilot program in Mexico for Centers of Excellence (August 18, 2026); UNESCO AI ethics and governance observatory; UN Global Dialogue on AI Governance coordinated by UNESCO and ITU.
Source for NIST AI data center architecture and security standards workshop (July 22-23); NIST AI risk management mission and AI RMF operationalization; NIST joining National Genesis Mission to accelerate AI innovation.
AI Regulation & Policyを毎週、自動で監視
このレポートは一次ソースのみから生成されています。テーマとソースを選べば、引用付きレポートが毎週届きます。7日間無料トライアル・$33/月から。
無料トライアルを始める