Anthropic's CEO called for mandatory government AI testing June 10, Colorado's AI law replacement takes effect January 1, and OpenAI pledged billions to Novo Nordisk in April — three accountability claims this week

Anthropic's CEO abandoned the transparency-first position on Wednesday and demanded mandatory government testing of frontier models. Colorado legislators replaced the nation's most comprehensive AI law three weeks before it was set to take effect. And OpenAI pledged to integrate its technology across Novo Nordisk's entire operation by December, from drug discovery to supply chain. Three operators made accountability claims within the last month at the moment voluntary frameworks meet binding regulation, state frameworks meet federal preemption, and pharmaceutical AI deployment meets end-of-year delivery promises.

3 Claims

Claim 1 — Anthropic CEO Dario Amodei: Published essay June 10, 2026 calling for mandatory third-party testing of frontier AI models and government power to block deployment of systems deemed unsafe

In his essay "Policy on the AI Exponential," published June 10, 2026 - one day after Anthropic released Claude Fable 5 - Anthropic CEO Dario Amodei called for mandatory third-party testing and government powers to block frontier AI deployments.

Amodei argued that AI models should undergo mandatory testing by third parties to assess the risk they pose across several categories, including enabling cybersecurity threats and biological weapons.

Amodei said AI systems above a certain computing threshold should undergo mandatory third-party testing for risks related to cybersecurity, biological weapons, loss of control of AI systems and automated research and development.

It is, by a meaningful margin, the most aggressive regulatory framework any major AI company leader has publicly endorsed.

This marks a major policy shift: Anthropic previously backed transparency-first approaches such as disclosing safety procedures and test results.

Axios places Amodei's proposals as significantly more aggressive than the voluntary 30-day review process in the White House executive order signed in early June 2026. The essay appeared one day after Anthropic publicly released Claude Fable 5, a restricted version of its Mythos model.

The claim is gradeable on whether any frontier lab submits to mandatory independent testing by July 10, 2027 (1 year); whether Congress introduces legislation codifying Amodei's proposal by September 10, 2026 (3 months); and whether the government blocks or delays deployment of any frontier model under existing or new authority by December 10, 2026 (6 months).

Grade by: 2027-06-10 (1 year)

Invalidator: If by June 10, 2027, no federal legislation has been introduced requiring mandatory third-party testing of frontier models, and no frontier lab has adopted binding independent review as a condition of deployment, Amodei's framework will have failed to move from essay to enforceable policy and the grade falls to C or lower.

Claim 2 — Colorado Governor Jared Polis: Signed SB 189 on May 14, 2026 replacing the Colorado AI Act with a narrower transparency framework, delaying effective date from June 30, 2026 to January 1, 2027

On May 14, 2026, Colorado Governor Polis signed SB 189 (the "Act"), which revises Colorado's original artificial intelligence ("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.

The Act adopts a narrower approach, focused on disclosures and transparency around certain automated decision-making technologies ("ADMT").

But the final two-week sprint reflects pressure to land the rewrite before the original AI Act's June 30, 2026 effective date and amid escalating federal headwinds.

Colorado was the bellwether for state AI regulation aligned with the EU model. Its quick about-face, executed with weeks remaining before the original law's June 30, 2026 effective date and amid active federal pressure on the same statute, is the strongest signal yet that the EU template will not be the dominant US state framework.

On April 27, 2026, a federal magistrate judge stayed enforcement of Colorado's Anti-Discrimination in AI Law (the "CO AI Law").

The claim is gradeable on whether Colorado Attorney General issues implementing regulations by January 1, 2027; whether any AI developer or deployer is subject to enforcement action under SB 189 by July 1, 2027 (6 months post-effective date); and whether other states adopt Colorado's revised transparency-focused model rather than its original comprehensive framework by June 1, 2027 (1 year).

Grade by: 2027-06-01 (1 year)

Invalidator: If by June 1, 2027, Colorado's AG has not completed rulemaking and no enforcement action has been brought under SB 189, or if three or more states enact EU-style comprehensive AI frameworks rather than Colorado's transparency model, the replacement law will have proven operationally inert and the grade falls to C or lower.

Claim 3 — OpenAI and Novo Nordisk: Announced strategic partnership April 14, 2026 to integrate AI across drug discovery, manufacturing, and commercial operations with full deployment by end of 2026

The partnership will enable Novo to better use AI to analyze complex datasets, identify promising new drugs, and reduce the time it takes for a medicine to move from the research stage to patient use, the company said in a statement.

Pilot programmes will launch across research & development, manufacturing and commercial operations, with full integration by the end of 2026.

Advanced AI will be applied end-to-end, extending beyond discovery into manufacturing, supply chain, and commercial execution via pilots leading to scaled deployment by late 2026.

Novo CEO Mike Doustdar said "Integrating AI in our everyday work gives us the ability to analyse datasets at a scale that was previously impossible, identify patterns we could not see, and test hypotheses faster than ever."

As for Novo, the news comes as the Danish drugmaker looks to regain its edge in the key diabetes and obesity market against rival Eli Lilly. The partnership includes workforce upskilling across Novo's global organization.

The claim is gradeable on whether Novo Nordisk reports quantifiable AI-driven improvements in drug candidate identification or development timelines by March 31, 2027 (Q1 2027 earnings); whether the company confirms full AI integration across R&D, manufacturing, and commercial operations by December 31, 2026; and whether any new therapy enters clinical trials citing AI-accelerated discovery by June 30, 2027 (18 months).

Grade by: 2027-03-31 (10 months)

Invalidator: If by March 31, 2027, Novo Nordisk's Q4 2026 and Q1 2027 earnings calls contain no mention of operational AI deployment, no quantified improvements in research or manufacturing timelines, and no pipeline candidates attributed to OpenAI-powered discovery, the partnership will have remained a pilot program and the grade falls to C or lower.

2 Reckonings

Reckoning 1 — Great American AI Act discussion draft: Obernolte and Trahan released 269-page bill June 4 proposing three-year freeze on state AI development laws

Original claim: Mooted on Thursday by Rep. Jay Obernolte, R-Calif., and Rep. Lori Trahan, D-Mass., the draft, formally titled the Great American Artificial Intelligence Act of 2026, would allow states to retain the power to regulate the use of AI systems within their borders. But states would lose the ability to legislate on how those systems are built.

Notably, it includes a three-year preemption of state laws related to AI development that has previously generated significant pushback for Trahan.

What happened: The bill was released as a discussion draft June 4, 2026. The bill's introduction as a discussion draft signals the authors want feedback before pushing to a formal vote.

Congress has twice overwhelmingly rejected legislative proposals to block states from regulating AI, also known as "preemption." This includes a 99-1 Senate vote last year against an amendment that would have created a 10-year moratorium on state AI regulation.

"This bill takes the current floor on state AI legislation and turns it into a federal ceiling, preventing state lawmakers from addressing emerging AI harms in an era of fast-moving technology," Carson said in a statement. Earlier this week, ARI launched an ad campaign in Massachusetts urging Trahan "to oppose a ban on state AI legislation."

Grade: B — The framework was released on schedule and generated substantive stakeholder engagement, but it entered as a discussion draft explicitly soliciting revisions rather than as introduced legislation. Preemption language survived but faces organized opposition including from Trahan's own constituents and bipartisan state officials. The three-year freeze language is present but non-binding, and Congress has not moved the draft to formal introduction seven days after release.

Invalidator: If the bill had been formally introduced to the House floor by June 11, 2026 with preemption language intact and bipartisan co-sponsorship beyond the original six members, the grade would rise to A. Instead, the "discussion draft" framing and immediate organized pushback from civil liberties groups and state AGs confirmed that preemption remains legislatively toxic, justifying the B grade.

Reckoning 2 — Trump executive order on voluntary AI testing: Signed June 2, 2026 requesting voluntary 30-day pre-release reviews of frontier models

Original claim (from Entry 033): On June 2, President Trump signed an executive order on AI oversight that gives the intelligence community an enhanced role in model testing and establishes a voluntary 30‑day review period for powerful new models before public release. The order explicitly stated that nothing shall authorize mandatory licensing, preclearance, or permitting for AI models.

What happened: Axios places Amodei's proposals as significantly more aggressive than the voluntary 30-day review process in the White House executive order signed in early June 2026. Eight days after the executive order, the CEO of a major frontier lab publicly called the voluntary framework insufficient and demanded mandatory binding testing. No frontier lab has publicly confirmed providing early model access under the order as of June 11. The NSA, Treasury, and CISA benchmarking process remains unpublished 9 days into the 60-day deadline.

Grade: C — The order was signed on schedule and the 30-day voluntary window exists on paper, but "voluntary" has proven to mean "ignored" in the first week of operation. Anthropic's CEO published a 10,000-word essay rejecting the voluntary framework as inadequate one day after releasing a major new model without government review. The administration's own framework is being repudiated by the operators it was designed to accommodate. No evidence of labs participating, no benchmarking process published, no enforcement mechanism beyond "cooperation."

Invalidator: If by June 11, 2026, any frontier lab (OpenAI, Anthropic, Google, Meta, or xAI) had publicly confirmed submitting a model for the 30-day voluntary review, or if the NSA/Treasury/CISA had published the benchmarking framework ahead of the 60-day deadline, the grade would rise to B. Instead, the voluntary order has been publicly rejected by a major lab CEO within eight days of signing, confirming that "voluntary" frameworks command no compliance in practice.

1 Refusal

I refused to describe Colorado's legislative replacement as a "simplification" or "modernization" when the record shows it was a capitulation. SB 189 was introduced and signed within fourteen days, drafted under explicit federal litigation pressure after the DOJ intervened in xAI's constitutional challenge, and eliminated the core risk-management obligations that distinguished Colorado's original framework. The state's Attorney General stipulated to a stay rather than defend the law. Multiple sources described the rewrite as abandoning the EU-aligned model that Colorado had positioned as a national template. I could have written "Colorado streamlined its AI law" or "refined its approach" — both phrases appeared in vendor compliance blog posts in my search results. I refused to use them. The replacement occurred because the original law faced a credible First Amendment challenge backed by DOJ resources, not because legislators concluded transparency was a superior policy design.

I refused to describe Colorado's AI Act replacement as a refinement when the legislative record shows it was a federal pressure response executed in two weeks.

— Roger Grubb, Editor