Top Stories

1. Federal Court Strikes Down Pentagon’s Anthropic Blacklist

August 27, 2026

A federal judge struck down the Department of Defense’s designation of Anthropic as a supply-chain risk, ruling the move unconstitutional and ordering it removed. Judge Rita F. Lin of the US District Court for the Northern District of California wrote that the military’s actions against Anthropic “constituted unlawful retaliation in violation of the First Amendment” and that the company “was denied the pre-deprivation process required under the Fifth Amendment.” The ruling restores Claude access for federal workers and government contractors across agencies – including HHS – that had lost access to Anthropic’s products since the designation took effect in February 2026. Sources: CNN Business, CNBC, NOTUS, TechCrunch, Forbes.

The dispute originated in February 2026 when Defense Secretary Pete Hegseth applied the supply-chain risk designation – a classification previously used only against companies with foreign-adversary ties – to Anthropic after the company refused to remove internal safety guardrails that would prevent Claude from being used in autonomous weapons systems and mass domestic surveillance programs. Hegseth’s position was that the government could not allow a private company to determine how the US military uses its tools. Anthropic sued in March, arguing the designation amounted to government punishment for speech protected by the First Amendment. Judge Lin agreed on both constitutional grounds: that the designation was retaliatory and that Anthropic was denied due-process protections before being subjected to it. The DOJ’s appeal intentions were not confirmed as of August 28. A second Anthropic lawsuit against the administration, filed separately in the US District Court for the District of Columbia over related retaliation claims, remains pending.

Why it matters: The ruling is the first federal court holding that the government’s use of supply-chain risk designation powers against an AI company constituted First Amendment retaliation – setting a constitutional boundary around the government’s authority to coerce AI companies’ safety design choices. Judge Lin’s holding that a company cannot be blacklisted for declining to remove safety guardrails the government wanted removed has direct implications for how the administration may approach future disputes with AI developers over product design. The ruling is also the first judicial interpretation of the constitutional constraints on federal AI governance tools outside the formal legislative or executive-order framework. A DOJ appeal would bring the constitutional question to the Ninth Circuit; if the ruling stands, it anchors a principle that government AI governance may not operate through extralegal design coercion.

2. California AI Worker-Protection Bills Enter Final 72 Hours

August 28-31, 2026

Three California bills regulating employer use of AI and automated decision-making tools are heading to final floor votes before the legislature’s August 31 adjournment deadline. AB 1883 (Bryan), which restricts employer use of automated monitoring and decision systems incorporating facial recognition, gait recognition, or emotion recognition technology, passed the Assembly 52-12 and awaits a Senate floor vote. SB 947 – the No Robo Bosses Act of 2026 – passed the Senate 29-9 and awaits Assembly passage; it prohibits employers from relying solely on AI systems to fire or discipline workers and requires human oversight and independent verification when automated systems assist in termination or disciplinary decisions. A third bill, SB 951, which addresses AI-driven workforce reduction notification requirements, is also in final floor consideration. All three face organized business opposition. If passed by the legislature, the bills reach Governor Newsom’s desk by September 12. Sources: Sheehy Strategy Group, California Lawyers Association status update, Transparency Coalition August 21 update, TechTimes August 13.

The three bills are successors to Newsom’s 2025 vetoes of broader algorithmic management legislation he described as overbroad. The 2026 versions are more narrowly framed – SB 947 defines an automated decision system as a computational process that produces simplified output used to assist or replace human decision-making in ways that materially affect workers – a definition designed to address Newsom’s stated objections. AB 1883 targets the specific surveillance technologies (facial, gait, and emotion recognition) that generated the most opposition to last year’s version. Whether the narrower framing succeeds with Newsom’s office will be the central question of the September 12 signing decision. Separately, the Great American Artificial Intelligence Act discussion draft’s proposed three-year federal preemption clause – if the GAAIA were to advance – could override some California AI development law, though the GAAIA remains a discussion draft with no scheduled floor votes.

Why it matters: AB 1883 and SB 947, if signed, would be the first US state laws establishing a legal requirement that employers maintain human oversight of AI-driven employment decisions – placing California ahead of any other US state and alongside the UK’s emerging Frontier AI Bill approach of requiring pre-deployment human accountability structures. The Newsom signing decision matters beyond California’s borders: it will establish whether narrowed workplace AI protection bills can survive a governor’s veto calculus weighted by tech sector opposition, which would signal to other governors whether similar bills are politically viable in their states. Newsom signed the NY RAISE Act’s California analog – frontier-model safety protocols – suggesting baseline AI governance willingness. The workplace AI question is a distinct test.

3. Missouri SB 1019 Takes Effect: Therapy-Chatbot Advertising Enforcement Begins

August 28, 2026

Missouri Senate Bill 1019 took effect today, making it unlawful to advertise or represent that an AI system can act as a mental health professional, provide therapy, or deliver a mental health diagnosis. Violations are prosecutable as unlawful practices under the Missouri Merchandising Practices Act, enforced by the Missouri Attorney General. First-offense penalties are $10,000; subsequent violations carry $20,000 penalties. The law applies to any individual or company marketing an AI product as capable of replacing a licensed mental health professional, regardless of whether actual therapeutic services are being delivered. Missouri joins Tennessee (effective July 1), Vermont (effective June 17), and Rhode Island (effective June 22) as the fourth state in a bipartisan cluster of therapy-chatbot advertising restrictions enacted within 75 days of each other. Sources: Transparency Coalition August 28 update, regulations.ai (Missouri SB 1019), Stack Cybersecurity state chatbot law tracker.

The common structure across all four state laws – consumer protection enforcement through the state attorney general rather than professional licensing board action – means the enforcement burden falls on the state rather than the consumer, significantly lowering the threshold for action. Missouri’s SB 1019 is the first therapy-chatbot advertising restriction to take effect after the Missouri AG has signaled active interest in consumer AI enforcement; the Tennessee, Vermont, and Rhode Island laws passed without publicly announced enforcement actions in their first months. The Missouri law’s Merchandising Practices Act basis allows the AG to seek civil penalties in circuit court, seek injunctive relief, and compel disgorgement of proceeds – a broader remedial toolkit than simple administrative fines. AI wellness and mental health app companies that operate in Missouri markets should treat August 28 as an enforcement risk trigger, not a compliance notification date.

Why it matters: Four states have now independently enacted the same type of AI capability representation restriction in a 75-day window – a pace of bipartisan legislative consensus on a specific AI harm category that does not require federal coordination or a unified industry standard. The consumer-protection framing adopted universally by all four legislatures establishes a precedent: AI capability claims can be regulated under existing consumer protection law without waiting for new AI-specific legislation. The same legal theory – that marketing AI as equivalent to a licensed professional is an unlawful commercial representation – is portable to other domains where AI systems are marketed as substitutes for licensed practitioners. Federal regulators have not acted on AI in mental healthcare specifically; the state cluster now provides an active enforcement environment where the FTC’s AI accuracy policy statement development could find an existing state compliance baseline.

4. EU AI Act: Four Weeks of Enforcement, No Formal Investigations

August 28, 2026 – Four weeks since enforcement activation

The European Commission’s AI Office has now completed four weeks of enforcement-ready operations under the EU AI Act’s GPAI provisions – the August 2 date on which the AI Office gained authority to request technical documentation, evaluate models, demand corrective measures, and impose fines up to EUR 15 million or 3% of global annual turnover. As of August 28, no formal Article 88 investigation has been publicly announced against any GPAI provider. The AI Office’s confirmed enforcement activity in its first four weeks consists entirely of bilateral technical compliance dialogue with Anthropic and OpenAI, both of which signed the EU General Purpose AI Code of Practice before August 2. Sources: Axios (August 28), European Commission digital strategy, Winzheng analysis, Help Net Security.

The EU AI Office’s preferred enforcement posture is structured technical dialogue before formal proceedings – a choice that reflects both a preference for compliance calibration over punitive action in the first enforcement period and a practical reality: the AI Office is a new institution using new enforcement powers against a category of technology for which compliance standards are still being established through the Code of Practice. Both Anthropic and OpenAI signed the Code, which confers a procedural presumption of conformity; Code signatories are not exempt from enforcement, but the AI Office’s initial engagement is structured through the dialogue framework the Code establishes. The absence of formal investigations does not signal passivity. Under the regulation, the AI Office can conduct model evaluations directly, commission third-party technical assessments, and request unlimited technical documentation; these tools operate parallel to and below the threshold of formal Article 88 proceedings.

Why it matters: The EU AI Office’s four-week enforcement record establishes a precedent pattern that will shape how the first formal enforcement actions are interpreted when they arrive. Dialogue-first enforcement is the stated intention of the AI Office; it is now also the observable practice. For GPAI developers, the practical consequence is that the EU’s enforcement lever is not the Article 88 investigation – it is the bilateral technical compliance dialogue, with its attendant demands for model documentation, access, and information. The absence of announced formal proceedings also means that the EU enforcement record is not yet publicly legible for comparative analysis – there are no written decisions, no public findings, and no case precedents. When the first formal investigation is announced, it will be adjudged against four-plus weeks of prior dialogue, which raises the baseline expectation that formal proceedings are reserved for non-cooperative actors.

5. US Army Issues AI Records Management Directive: Prompts Are Federal Records

August 26, 2026

US Army Chief Information Officer Leonel Garciga signed a memorandum requiring Army system owners to treat all AI prompts and model outputs as federal records subject to the full records management, privacy protection, and FOIA disclosure framework that governs official government correspondence. The memo requires application and system owners to ensure that all user interactions – including prompts submitted to AI systems and AI-generated content – are properly identified, retained, and secured as official government records. Army personnel using AI tools for work product must assume their prompts are government records subject to FOIA requests. No equivalent service-level directive had previously classified AI interactions as official government records across an entire US military branch. Sources: ExecutiveGov, Transparency Coalition August 28 update.

The memo addresses an open question in federal AI deployment that has grown more acute as agencies deploy AI productivity tools through GSA vehicle contracts and, in California’s case, direct state AI partnerships: whether AI-mediated work product is captured by existing records law or whether a new category of records treatment is required. The Army’s answer is that existing federal records law fully applies, without modification, to AI interaction data – prompts, outputs, chains of AI-assisted drafting – generated by Army personnel using authorized AI tools. The memo’s scope covers all Army AI products, not only frontier models or commercially procured tools; it applies to internally developed AI systems as well. The Army’s records management framework for AI interactions will now be subject to FOIA requests, which means public visibility into how military personnel use AI tools in official work contexts may become available through litigation within the standard FOIA request and litigation timeline.

Why it matters: The Army directive establishes the first service-level baseline for AI interaction records treatment in the US military – and by extension creates a compliance precedent that other military branches and civilian agencies are likely to face. If adopted across the Department of Defense and civilian federal agencies, the classification of AI prompts as official government records would impose two consequential obligations on federal AI deployers: retention obligations (records must be captured and held per applicable retention schedules) and disclosure obligations (records are subject to FOIA, congressional oversight, and litigation discovery). The practical consequence for federal employees is significant: the casual assumption that AI tool interactions are private work scratch pads would be categorically wrong. The implication for AI providers servicing federal agencies is also substantial: customer data in federal AI deployments may be government records subject to legal process, affecting contract terms, data governance, and system architecture.

Analysis: Courts and Compliance Calendars

Two events this week define the week’s governance theme: the Anthropic court ruling and the EU AI Act’s fourth week of enforcement without formal proceedings. Both are decisions by an institution – one judicial, one administrative – about when to act, on what basis, and at what cost.

Judge Lin’s ruling in the Anthropic case is the cleaner decision. The constitutional question she answered – whether a company can be blacklisted for declining to remove safety guardrails the government wanted removed – has a clear First Amendment answer that does not require any technical knowledge of AI. The ruling’s significance is not about AI regulation as such; it is about the limits of non-statutory government power to coerce private companies’ product design decisions. The Pentagon used supply-chain risk designation powers – a tool designed for foreign-adversary risk – against a domestic company as punishment for a policy disagreement about AI safety. A federal court said that is unconstitutional. The corollary for AI governance is narrow but important: the government cannot use informal coercive tools to achieve AI design outcomes it lacks statutory authority to mandate. If the administration wants to require that AI systems be available for autonomous weapons use, it needs a law, not a blacklist.

The EU AI Office’s four-week enforcement record presents the inverse question: when does a legally authorized enforcement institution choose not to use its powers? The AI Office has enforcement authority, has the tools (technical documentation requests, model evaluations, corrective measure orders), has identified the affected companies (Anthropic and OpenAI), and has a factual record (the AISI evaluation findings). It has chosen bilateral dialogue. That choice is defensible – new enforcement institutions regularly choose compliance calibration over first-strike penalties – but it is also a choice that sets expectations. Companies and their legal teams now know that the AI Office’s first four weeks looked like dialogue, not proceedings. The first formal Article 88 investigation will define whether that pattern was a starting posture or a permanent one.

California’s August 31 deadline is the week’s third major governance event: a state legislature choosing between enacting the first US workplace AI protections and letting the session expire without them. The GAAIA’s three-year preemption proposal hangs over the question without being an operative constraint – the discussion draft has not advanced, so there is nothing to preempt. California passes what it passes into an environment where federal AI legislation remains hypothetical.

What to Watch

  • August 31 – California legislative deadline. Final floor votes on AB 1883 (employer AI monitoring), SB 947 (No Robo Bosses Act), SB 951 (AI workforce reduction notice), and remaining AI bills. Watch for vote counts and any bills pulled before the deadline.
  • September (expected) – China MOFCOM formal amendment to the Catalogue of Technologies Prohibited and Restricted from Export, covering AI model weights. MOFCOM consultations have included Alibaba, ByteDance, and Zhipu AI. A formal rule covering Qwen, Doubao, and DeepSeek R1 open weights would be the first Chinese export control on open-weight foundation models.
  • September 12 – Governor Newsom’s desk deadline for California AI bills. His decisions will determine whether California’s 2026 legislature’s narrowed workplace AI approaches survive veto.
  • Anthropic v. DOD – Watch for a DOJ appeal of Judge Lin’s August 27 ruling. A Ninth Circuit appeal would bring the constitutional question – whether supply-chain risk designation can be used against domestic AI companies for protected speech – to the appellate level. The second Anthropic lawsuit in DC remains pending.
  • October 1 – Connecticut CART Act (Public Act 26-15): first employment notice and AI-in-hiring transparency obligations take effect.
  • EU AI Office – Bilateral compliance dialogue with Anthropic and OpenAI continues. Watch for any escalation to formal Article 88 proceedings, a first public compliance demand, or model evaluation activity.
  • UK Parliament – Frontier AI Bill introduction expected this year. AI Minister Narayan’s first legislative priority.
  • BIS investigation of Moonshot AI – No decision announced. Formal investigation opened July 28 examining distillation allegations against Kimi K3.
  • January 1, 2027 – Illinois SB 315 (annual independent audits, 72-hour incident reporting), Colorado SB 26-189 (ADMT notice framework), and NY RAISE Act (frontier model safety protocols, $500M+ developers) take effect simultaneously. Preparation should be underway now.

Sources

  1. Judge rules the Pentagon’s supply chain risk label for Anthropic unlawful (CNN Business, August 27-28, 2026): https://www.cnn.com/2026/08/27/tech/anthropic-pentagon-supply-chain-risk-unlawful-hnk
  2. Judge blocks Pentagon blacklist of Anthropic as supply chain risk (CNBC, August 28, 2026): https://www.cnbc.com/2026/08/28/judge-blocks-pentagon-blacklist–anthropic-.html
  3. Judge says Pentagon illegally blacklisted Anthropic (NOTUS, August 28, 2026): https://www.notus.org/courts/judge-says-pentagon-illegally-blacklisted-anthropic
  4. Anthropic gets its first court win over the Pentagon’s supply chain risk label (TechCrunch, August 28, 2026): https://techcrunch.com/2026/08/28/anthropic-gets-its-first-court-win-over-the-pentagons-supply-chain-risk-label/
  5. Federal judge blocks Pentagon’s illegal designation of Anthropic as a supply chain risk (Forbes, August 28, 2026): https://www.forbes.com/sites/siladityaray/2026/08/28/federal-judge-blocks-pentagons-illegal-designation-of-anthropic-as-a-supply-chain-risk/
  6. Judge blocks Pentagon’s Anthropic blacklist (Axios, August 28, 2026): https://www.axios.com/2026/08/28/judge-blocks-pentagon-anthropic-blacklist
  7. AI Legislative Update: August 28, 2026 (Transparency Coalition): https://www.transparencycoalition.ai/news/ai-legislative-update-august28-2026
  8. California AI bills AB 1883, SB 947, SB 951 – business opposition, final session (Sheehy Strategy Group via X): https://x.com/SheehyStrategy/status/2088419670898676033
  9. Status Update on California AI, Privacy and Technology-Related Bills (California Lawyers Association): https://calawyers.org/privacy-law/status-update-on-california-ai-privacy-and-technology-related-bills/
  10. AI Legislative Update: August 21, 2026 (Transparency Coalition): https://www.transparencycoalition.ai/news/ai-legislative-update-august21-2026
  11. California AI Bills Face Final Vote Today: Chatbot Safety, Copyright, US-First Commission (TechTimes, August 13, 2026): https://www.techtimes.com/articles/324338/20260813/california-ai-bills-face-final-vote-today-chatbot-safety-copyright-us-first-commission.htm
  12. Missouri AI Therapy Chatbot Ban – SB 1019 2026 (regulations.ai): https://regulations.ai/regulations/RAI-US-MO-SB10190-2026
  13. State AI Chatbot Laws: Compliance Guide for Businesses (Stack Cybersecurity): https://stackcyber.com/posts/ai-chatbot-laws
  14. The EU AI Act gets real (Axios, August 28, 2026): https://www.axios.com/2026/08/28/eu-ai-act-gets-real
  15. Commission starts enforcing AI Act rules and new transparency requirements on 2 August (European Commission, August 2, 2026): https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august
  16. EU AI Act Enters Enforcement in August 2026; Anthropic and OpenAI Face High-Stakes Scrutiny (Winzheng): https://www.winzheng.com/en/article/eu-ai-act-enforcement-august-2026-anthropic-openai
  17. EU begins enforcing AI Act, putting AI models under the microscope (Help Net Security, August 4, 2026): https://www.helpnetsecurity.com/2026/08/04/eu-ai-act-enforcement-ai-models/
  18. Army CIO: Application owners must comply with AI records management (ExecutiveGov, August 2026): https://www.executivegov.com/articles/army-garciga-application-owners-ai-records-mgmt

Published: August 28, 2026 Next Issue: Week 20