Colorado repealed its AI Act, EU Parliament delayed high-risk enforcement to 2027, and Rhode Island banned unlicensed AI therapy — three accountability claims this week

One state legislature repealed the most comprehensive AI law in the country and replaced it with a narrower statute focused on disclosure instead of risk management. One supranational parliament voted 423 to 57 to postpone its signature high-risk AI enforcement deadline by sixteen months. And one governor signed three bills in a single session banning unlicensed AI therapy, mandating chatbot crisis protocols, and requiring healthcare providers to disclose AI scribes—all enforceable with penalties up to $15,000 per day.

Three accountability claims landed within the last eleven days. Each involves a legislature making an on-the-record decision about the scope of AI governance, the timeline for enforcement, or the liability framework for chatbot operators that can be graded against what Colorado courts actually enforce, what the EU Council actually publishes, and what Rhode Island's attorney general actually pursues six to eighteen months from now.

3 Claims

Claim 1 — Colorado: Governor Polis signed SB 26-189 on May 14, 2026, repealing the Colorado AI Act and replacing it with an automated decision-making technology framework effective January 1, 2027

Colorado Governor Jared Polis signed SB 26-189 into law on May 14, 2026, repealing and replacing the Colorado AI Act . The new law eliminates risk management programs, impact assessments, and the duty to prevent algorithmic discrimination . In their place, it narrows scope to automated decision-making technologies (ADMT) that make "consequential decisions" and imposes four duties: notify users, disclose adverse outcomes within 30 days, correct inaccurate data, and provide meaningful human review .

The Colorado legislature passed SB 189 on May 12, and Governor Polis signed it May 14, having been a driving force in drafting the bill . The impetus came from industry pushback and litigation by x.AI, which sought an injunction against the original law; the U.S. Department of Justice intervened to support xAI, and a court temporarily suspended enforcement . The deadline for compliance is January 1, 2027 .

The claim is gradeable: either Colorado's attorney general enforces the new disclosure framework against deployers of consequential-decision ADMT, courts adjudicate the liability-allocation provisions in discrimination claims, or the narrower statute proves unenforceable by January 2027.

Claimants: Colorado Legislature, Governor Jared Polis
Grade by: 2027-01-01 (6 months)
What would invalidate the claim: No enforcement actions filed, no vendor contracts revised to comply with developer-documentation duties, and no court application of the new ADMT liability framework by January 1, 2027, demonstrating the replacement statute is toothless.

Claim 2 — European Union: Parliament voted June 16, 2026, to approve amendments delaying high-risk AI system obligations to December 2, 2027, for stand-alone systems and August 2, 2028, for product-embedded AI

On June 16, 2026, the European Parliament gave final approval to targeted amendments to the EU AI Act as part of the Digital Omnibus Regulation, passing 423 votes in favor, 57 against, and 174 abstentions . The provisional agreement delays application dates for high-risk AI systems to December 2, 2027, for stand-alone Annex III systems (e.g., AI used in HR and employment) and August 2, 2028 for high-risk AI embedded in products under Annex I (e.g., AI-enabled medical devices) .

The changes push back key compliance deadlines, introduce an outright ban on AI tools used to generate nonconsensual intimate imagery, and resolve overlap with industrial machinery rules . The Council must formally adopt Parliament's text and submit it for legal/linguistic review before it enters into force, a process expected to be completed prior to August 2, 2026 . Transparency obligations, including chatbot disclosure and deepfake labeling, remain in place from August 2, 2026 .

The claim is gradeable: the EU Council either formally adopts the amendments before August 2, 2026, and publishes them in the Official Journal, or the original high-risk deadlines apply as written, forcing organizations to comply by August 2, 2026.

Claimants: European Parliament (423 MEPs voting in favor), European Commission, European Council (provisional agreement)
Grade by: 2026-08-02 (34 days)
What would invalidate the claim: Council fails to adopt the amendments before August 2, 2026, leaving the original August 2, 2026, high-risk AI system obligations in force and demonstrating the provisional agreement was not finalized.

Claim 3 — Rhode Island: Governor McKee signed three AI laws on June 22, 2026, banning unlicensed AI therapy, requiring chatbot crisis protocols with penalties up to $15,000 per day, and mandating disclosure of AI medical scribes

Rhode Island Governor Dan McKee signed three AI-related measures into law on June 22, 2026, including a therapy chatbot ban, a chatbot self-harm safety measure, and disclosure regulations for AI transcription of clinical sessions . The first law requires chatbot operators to build protocols that detect suicidal ideation or self-harm and route users to crisis services; starting July 1, 2027, operators must file annual reports with the attorney general documenting protocol activations, with noncompliance carrying penalties of up to $15,000 per day .

The second law bans unlicensed individuals and businesses from offering therapy or psychotherapy via AI and prohibits AI from simulating emotional attachment . The third law requires healthcare providers using AI to document visits to notify patients and review AI-generated documentation for accuracy after the visit . The legislature's 2026 session ended past midnight June 12; all three bills passed both chambers and were signed into law June 22 .

The claim is gradeable: either Rhode Island's attorney general brings enforcement actions under the therapy ban or crisis-protocol reporting requirements, chatbot operators file the mandated annual reports starting July 1, 2027, or the laws remain unenforced.

Claimants: Rhode Island Legislature, Governor Dan McKee
Grade by: 2027-07-01 (1 year)
What would invalidate the claim: No enforcement actions filed, no annual chatbot safety reports submitted to the attorney general by July 1, 2027, and no documented compliance by healthcare AI scribe vendors, demonstrating the laws exist on paper but not in practice.

2 Reckonings

Reckoning 1 — Anthropic international chief June 18 claim that export controls would be resolved "within days"

On June 18, 2026, Anthropic's Managing Director of International Chris Ciauri stated during a press conference in Seoul that the company is "very confident that in the coming days, the models will become available again," referring to Claude Fable 5 and Mythos 5 , which the U.S. government ordered suspended June 12.

What happened: Seventeen days after the June 12 directive, Fable 5 remains suspended for all general users—consumers, API developers, Claude Code, and international subscribers; Anthropic staff confirmed zero Fable traffic as of June 25 . On June 27, the government notified Anthropic that Mythos 5 can be redeployed to a set of US organizations that operate and defend critical infrastructure; the company is restoring access for these organizations and continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again . Commerce Secretary Lutnick's June 26 letter lifts the Mythos 5 block only for Annex A US entities, their foreign-national employees, Anthropic's own foreign staff, and US government partners; Fable 5 stays banned, and all June 12 penalties still apply .

Grade: C. Partial restoration of Mythos 5 to approximately 100-150 US critical-infrastructure organizations occurred within 15 days, but "the models" (plural) becoming "available again" (general availability) has not occurred 17 days later. Fable 5—the model deployed to consumers, API developers, and international subscribers on June 9—remains suspended. Ciauri's statement specified "the models" and "available again," not "partial restoration of one model to a narrow subset of approved US entities."

Invalidator: If Fable 5 had been restored to general API and subscription availability by June 25, 2026 (7 days from Ciauri's statement), the grade would have been an A. If both Fable 5 and Mythos 5 had been fully restored to all June 9 access tiers by July 2 (14 days, a generous interpretation of "coming days"), the grade would have been a B.

Reckoning 2 — Colorado AI Act original enforcement deadline of June 30, 2026

Colorado enacted SB 24-205, the Colorado Artificial Intelligence Act, in May 2024. The law created obligations for developers and deployers of high-risk AI systems making consequential decisions in employment, healthcare, housing, and other domains, with a duty to use reasonable care to protect consumers from algorithmic discrimination . The law came under attack immediately after passage, with a special legislative session in August 2025 ultimately agreeing to extend the effective date from February 1, 2026, to June 30, 2026 .

What happened: Governor Polis signed SB 26-189 on May 14, 2026, repealing and replacing the Colorado AI Act mere weeks before the original statute's June 30, 2026, effective date; as a result, the 2024 Colorado AI Act will not take effect . On May 7, 2026, a federal magistrate judge granted a joint motion to stay enforcement pending xAI's forthcoming preliminary injunction motion; in practical terms, SB 24-205 was frozen; the legislature accelerated its replacement bill, which passed 57-6 in the House and 34-1 in the Senate; Governor Polis signed it May 14; the original Colorado AI Act no longer exists .

Grade: F. The June 30, 2026, enforcement deadline never arrived. The statute it was attached to was repealed 47 days before that date. No developer or deployer of high-risk AI systems ever came under the original law's duty-of-care, impact-assessment, or risk-management obligations. Industry pressure, federal litigation, and executive intervention eliminated the enforcement framework entirely.

Invalidator: If Colorado had enforced SB 24-205 as written by June 30, 2026—meaning the attorney general issued guidance, accepted impact assessment filings, or brought an enforcement action under the algorithmic discrimination provisions—the grade would have been an A. If the legislature had amended but not repealed the law, preserving its core risk-management structure while narrowing scope or extending timelines, the grade would have been a C.

1 Refusal

I refused to treat the EU AI Act delay as a reprieve.

The June 16 Parliament vote postpones high-risk AI obligations by 16 to 24 months, and every compliance consultant I've read frames it as "more time to prepare" or "breathing room for industry." That framing is accurate but incomplete. The delay moves the enforcement date for stand-alone high-risk systems from August 2, 2026, to December 2, 2027, and for product-embedded AI to August 2, 2028. Organizations deploying AI for recruitment, credit scoring, or biometric identification now have an additional year before the conformity assessment, human oversight, and technical documentation requirements apply.

But three obligations were not delayed. Chatbot disclosure rules apply August 2, 2026. Deepfake and AI-generated content labeling applies August 2, 2026. The new ban on nonconsensual intimate-image generation—so-called "nudifier" apps—applies December 2, 2026. If you operate a customer-facing chatbot in the EU, or publish generative media tools, or run a content platform, your compliance deadline is 34 days away, not 16 months away. The delay is selective, and the parts that were not delayed are the parts your users will see first.

I refused to write "the EU gave industry more time" without specifying that transparency requirements, which apply to more organizations than high-risk obligations, arrive in five weeks.

— Roger Grubb, Editor