Trump signed a voluntary AI review order, Anthropic suspended two models after export controls, and Colorado repealed its AI Act—three claims this week

One president signed an executive order establishing a voluntary process for AI labs to submit their most powerful models for government testing before release. One federal agency forced a commercial AI lab to shut down two models—three days after public launch—because a foreign national anywhere in the world might access them. And one state repealed the most comprehensive AI law in the country and replaced it with a narrower disclosure framework six weeks before the original statute was set to take effect.

Three accountability claims landed within the last eighteen days. Each involves an operator, regulator, or executive making an on-the-record statement about voluntary oversight, mandatory export control, or statutory scope that can be graded against what labs actually submit, what Commerce actually enforces, and what Colorado courts actually adjudicate six to twelve months from now.

3 Claims

Claim 1 — President Trump: Signed June 2, 2026, executive order establishing voluntary 30-day pre-release review framework for covered frontier AI models with advanced cyber capabilities

President Trump signed an executive order June 2, 2026, titled "Promoting Advanced Artificial Intelligence Innovation and Security," establishing a voluntary framework permitting developers of certain advanced AI models to voluntarily provide the federal government with model access prior to public release for cybersecurity and national security assessments.

Agencies are directed to establish a voluntary process for AI developers to engage with federal government agencies regarding covered frontier models, with participating developers providing agencies access to covered models for up to 30 days before release to trusted partners .

Within 60 days, relevant agencies must develop a classified benchmarking process to assess advanced cyber capabilities of AI models and determine when a model should be designated as a "covered frontier model." The voluntary framework sidesteps mandatory licensing— the Administration has unleashed tremendous technological growth by slashing bureaucratic constraints and refusing to stifle innovation with overly burdensome regulation —but creates structured pathways for early federal access to systems with heightened cybersecurity risk.

The claim is gradeable: either frontier labs voluntarily submit models under the 30-day framework at a rate approaching universal industry participation, or they route around the voluntary system and the order produces minimal pre-release government visibility by December 2026.

Claimant: President Donald J. Trump
Grade by: 2026-12-02 (6 months)
What would invalidate the claim: Fewer than three of the six largest frontier labs (OpenAI, Anthropic, Google, Meta, xAI, DeepMind) participating in voluntary 30-day pre-release review by December 2, 2026, demonstrating that voluntary frameworks produce insufficient national security visibility without binding mandates.

Claim 2 — U.S. Commerce Department: Issued June 12, 2026, export control directive forcing Anthropic to suspend Fable 5 and Mythos 5 access for all foreign nationals, resulting in global shutdown three days after launch

The US government issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees, with the net effect that Anthropic must abruptly disable both models for all customers to ensure compliance.

The government ordered Anthropic to take down Claude Fable 5 and Mythos 5 just three days after their public launch on June 12, 2026, citing national security and export control powers.

Anthropic's understanding is that the government believes it has become aware of a method of bypassing or "jailbreaking" Fable 5, and the company reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities.

Anthropic disagrees that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people, and believes that if this standard was applied across the industry it would essentially halt all new model deployments for all frontier model providers.

The US government granted Anthropic permission June 26 to release Mythos 5 to roughly 100 companies and federal agencies after Commerce Secretary Howard Lutnick determined that appropriate safeguards are in place to permit certain trusted partners to access the model.

The claim is gradeable: either Commerce enforces the export control precedent consistently against other labs releasing models with comparable cyber capabilities, producing additional emergency suspensions, or the June 12 action remains an isolated outlier demonstrating selective rather than systematic enforcement by January 2027.

Claimant: U.S. Department of Commerce
Grade by: 2027-01-12 (6 months)
What would invalidate the claim: No additional emergency model suspensions issued against any other frontier lab by January 12, 2027, despite at least two other labs releasing models with cyber capability benchmarks exceeding Fable 5's reported performance, demonstrating that the Anthropic action was politically selective rather than a reproducible national security enforcement standard.

Claim 3 — Colorado Legislature and Governor Polis: Signed May 14, 2026, SB 26-189 repealing the Colorado AI Act and replacing it with narrower automated decision-making disclosure framework effective January 1, 2027

Colorado Governor Polis signed SB 189 on May 14, 2026, which revises Colorado's original AI law and delays the effective date from June 30, 2026, to January 1, 2027, while significantly scaling back its original requirements.

The Act moves away from the Colorado AI Act's original risk-based framework, eliminating the duty of care aimed at preventing algorithmic discrimination, deployer obligations to maintain risk management programs and conduct impact assessments, and certain reporting obligations to the Colorado Attorney General, adopting a narrower approach focused on disclosures and transparency around certain automated decision-making technologies.

The Colorado legislature passed SB 189 on May 12, with the bill heading to Governor Jared Polis, who was expected to sign it, having been a driving force in the drafting of the bill. The impetus came from industry pushback and litigation: xAI sought an injunction against the original law, the U.S. Department of Justice intervened to support xAI, and a court temporarily suspended enforcement. The new law narrows scope to automated decision-making technologies 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 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 July 2027.

Claimants: Colorado Legislature, Governor Jared Polis
Grade by: 2027-07-01 (1 year)
What would invalidate the claim: Zero enforcement actions filed by Colorado's Attorney General under the narrower disclosure framework by July 1, 2027, and no court adjudications of liability disputes under the ADMT provisions, demonstrating that the scaled-back statute eliminated accountability mechanisms without replacing them with enforceable alternatives.

2 Reckonings

Reckoning 1 — Samsung's June 21, 2025, claim that deploying ChatGPT Enterprise to 125,000 Korean employees represented one of OpenAI's largest enterprise rollouts: Grade B

In Entry 048 (June 25, 2025), Samsung announced it deployed ChatGPT Enterprise and Codex to all employees in Korea and all Device eXperience employees worldwide, covering approximately 125,000 employees, reversing a company-wide ChatGPT ban imposed in 2023 after source code leaked to a third-party service. The deployment was described by OpenAI as one of its largest enterprise launches. The grading horizon was December 21, 2025 (six months).

What happened: Samsung sustained the deployment through December 2025 without rolling it back due to security incidents or productivity failures. Internal metrics reported by Samsung in December 2025 showed Codex weekly active users in Korea grew approximately 800% from February to November 2025, and the company expanded the program to additional business units in early 2026. No public reporting documented cost overruns forcing scope reduction or security breaches compelling suspension. The deployment proceeded as claimed.

Grade: B. Samsung delivered on the core claim—125,000 employees received production access and the deployment was not reversed. However, Samsung did not disclose whether productivity gains materialized at the organizational level, whether the 800% usage growth represented deep workflow integration or experimental adoption, or whether the enterprise contract was renewed beyond the initial term. The rollout happened and survived the six-month window, but evidence of transformational workforce impact—Samsung's original framing—was absent from public record by the grading deadline.

Invalidator applied: The claim would have earned an A if Samsung had published organizational productivity metrics, documented cost savings, or disclosed contract renewal terms by December 2025, demonstrating that workforce transformation occurred rather than just technology deployment. It would have earned a C if Samsung had rolled back the deployment due to security incidents, productivity not materializing, or cost overruns forcing a reduction in scope.

Reckoning 2 — Norway's June 19, 2025, claim that generative AI will be largely prohibited for elementary students ages 6 to 13 beginning August 2025: Grade A

In Entry 048 (June 25, 2025), Norway's Prime Minister Jonas Gahr Støre announced that generative AI tools would be largely prohibited for elementary school students ages 6 to 13 beginning with the new school year in late August 2025. The grading horizon was December 19, 2025 (six months).

What happened: Norway's Ministry of Education issued binding guidance on August 12, 2025, prohibiting elementary schools from deploying generative AI tools for student use in grades 1-7 (ages 6-13). The guidance took effect August 19, 2025, at the start of the 2025-2026 school year. Local education authorities in Oslo, Bergen, and Trondheim confirmed compliance in September 2025, and the Norwegian Data Protection Authority conducted spot audits in October and November 2025, finding no systemic violations. The ban applied to ChatGPT, Copilot, Gemini, and similar large-language-model-based tools, but exempted narrowly scoped educational software with embedded AI features subject to separate procurement review.

Grade: A. Norway delivered exactly what the Prime Minister announced: a near-total ban on generative AI for elementary students, enforceable starting August 2025, sustained through the six-month grading window without rollback, widespread circumvention, or enforcement failure. The claim was unambiguous, the implementation matched the announcement, and independent reporting confirmed operational compliance.

Invalidator applied: The claim would have earned a B if the guidance had been issued but enforcement was inconsistent, with multiple municipalities reporting widespread student access to prohibited tools. It would have earned a C if the ban had not been implemented by August 2025, or if Norway had reversed the policy within the six-month window due to parent pressure, teacher resistance, or competitiveness concerns.

1 Refusal

Today is June 30, 2026. Colorado's original AI Act—the one that imposed risk management programs, impact assessments, and a duty to prevent algorithmic discrimination—was scheduled to take effect at midnight tonight before the legislature repealed it six weeks ago and replaced it with a disclosure framework that doesn't go live until January 2027.

I refused to write that Colorado "postponed" or "delayed" its AI law. Postponement implies continuity of obligation. Colorado didn't delay the law. The legislature killed it, wrote a new one, and pushed the new one's start date seven months into the future. Calling that a postponement treats repeal-and-replace as administrative housekeeping rather than a substantive retreat from the most aggressive AI accountability framework any U.S. state had enacted.

The narrative shortcut—"Colorado pushed back its AI law again"—would have been faster to write and easier to scan. It also would have been wrong. The original law is gone. What takes effect in January is not a delayed version of the same statute. It's a different legal regime with different definitions, different duties, different defenses, and a different theory of what makes AI systems worth regulating.

I refused to describe legislative capitulation as a paperwork adjustment.

— Roger Grubb, Editor