Coinbase cut 700 jobs and called it AI. The government promised to test frontier models — but told no one what passing means. And Novo Nordisk bet its drug pipeline on full OpenAI integration by year's end.

A cryptocurrency exchange cut 14 percent of its workforce in a single day and called it "optimizing for the AI era." A government agency announced it would test unreleased AI models from three frontier labs—but published no criteria defining what a model must demonstrate to pass, fail, or deploy anyway. And a pharmaceutical giant committed to full AI integration across drug discovery, manufacturing, and commercial operations by December 31, with its CEO promising to "bring new and better treatment options to patients faster" than competitors spending billions on the same bet.

All three claims landed in the public record within the last 72 hours. All three involve operators who cannot quietly retract the timeline, the workforce number, or the deployment date without becoming the story. And all three can be graded six months from now against SEC filings, government disclosures, and the company's own earnings calls.

That is the job of this ledger.

3 Claims

Claim 1 — Coinbase: 700 layoffs (14% of workforce) to "rebuild the company as an intelligence"

On May 5, 2026, Coinbase announced it would cut 700 jobs, or roughly 14% of its workforce, as it shifts toward a more AI-centric workflow . CEO Brian Armstrong told employees in a companywide memo that the cuts were not simply cost reduction: the company is "fundamentally changing how we operate, rebuilding the company as an intelligence, with humans around the edge aligning it" .

Leaders at Coinbase are saying that in order to take advantage of the power of AI, the company needs to reduce headcount and realign its org chart so teams can use AI more productively . The claim is graded on whether Coinbase completes the 700 layoffs by May 31, 2026; whether the company frames AI productivity gains as the reason in subsequent earnings calls and SEC filings; and whether headcount stays below 4,300 through year-end 2026 absent acquisitions or disclosed rehires.

The invalidator would be credible reporting—via Coinbase investor relations, SEC filings, or investigative journalism from The Block, CoinDesk, or CBS News—showing that fewer than 600 employees were laid off, that Coinbase disclosed the cuts were primarily due to market conditions unrelated to AI, or that the company rebuilt the eliminated roles by Q4 2026 without acknowledging the reversal.

Grade by: 2026-11-07 (6 months)

Claim 2 — Commerce Department: Pre-deployment testing of Google, Microsoft, xAI models with no published pass/fail criteria

On May 5, 2026, the Center for AI Standards and Innovation announced agreements with Google DeepMind, Microsoft, and Elon Musk's xAI that will allow the U.S. government to evaluate artificial intelligence models before they are publicly available, conducting pre-deployment evaluations and targeted research to better assess frontier AI capabilities . The announcement builds on CAISI's previous partnerships with OpenAI and Anthropic from 2024, with those agreements renegotiated to reflect directives from Commerce Secretary Howard Lutnick and America's AI Action Plan .

The agreement allows CAISI to test models for national security risks before public launch. The center has already completed more than 40 AI model evaluations . But neither the Commerce Department nor CAISI has published criteria defining what constitutes a passing score, what risk level triggers a deployment delay, or whether companies can release models that fail review.

The claim is gradeable on whether CAISI publishes a transparency report by November 7, 2026, disclosing: (1) the number of models evaluated from Google, Microsoft, and xAI; (2) the pass/fail outcomes or risk ratings assigned to each; and (3) whether any model was delayed, modified, or denied deployment as a result of the evaluation. The invalidator would be CAISI or Commerce Department disclosure—via press release, congressional testimony, or investigative reporting from Axios, CNBC, or The New York Times—showing that no such transparency mechanism was established, or that all models tested were approved for deployment regardless of findings.

Grade by: 2026-11-07 (6 months)

Claim 3 — Novo Nordisk: Full OpenAI integration across operations by end of 2026

On April 14, 2026, Novo Nordisk announced a partnership with OpenAI to "bring new and better treatment options to patients faster," enabling the company to analyze complex datasets, identify promising new drugs, and reduce the time it takes for a medicine to move from research to patient use . Pilot programmes will launch across research & development, manufacturing and commercial operations, with full integration by the end of 2026 .

CEO Mike Doustdar framed the partnership as existential: Novo is racing Eli Lilly for obesity-drug dominance after Lilly surged ahead with its ZepBound regimen. The news comes as Novo looks to regain its edge in the diabetes and obesity market against rival Eli Lilly, following a 40% stock price decline, a CEO change, and wide-ranging layoffs .

The claim is gradeable on three dimensions: (1) whether Novo Nordisk discloses in its Q4 2026 or 2027 annual report that OpenAI tools have been deployed operationally across R&D, manufacturing, and commercial divisions by December 31, 2026; (2) whether the company quantifies productivity gains, timeline reductions, or cost savings attributable to the integration; and (3) whether investigative reporting or earnings-call transcripts reveal deployment delays, pilot failures, or scaled-back rollout. The invalidator would be Novo investor relations disclosure, earnings-call remarks, or credible reporting from FiercePharma, STAT, or Drug Discovery World showing that full integration was not completed by year-end, or that the partnership was quietly narrowed in scope without public acknowledgment.

Grade by: 2027-01-31 (1 year)

2 Reckonings

Reckoning 1 — FDA: AI integrated across all centers by June 30, 2025 (deadline passed 10 months ago)

On May 8, 2025, FDA Commissioner Martin Makary announced an aggressive timeline to scale use of artificial intelligence internally across all FDA centers by June 30, 2025, and directed all FDA centers to begin deployment immediately . The deadline was June 30, 2025—now 10 months in the past.

What happened: By March 2026, Makary stated that the FDA's AI tools had saved more than 14,000 staffers 17,000 hours of human work time since implementation at the end of June, suggesting the original June 2025 deadline was at least partially met . But in July 2025, CNN reported that FDA's AI system was "making up studies," raising questions about the quality and reliability of the deployment .

The FDA met the calendar deadline but delivered a system that fabricated citations—a failure mode the original claim did not anticipate but that an invalidator would capture. The agency has not published a post-deployment audit confirming which centers are using AI, which workflows were automated, or what share of the "17,000 hours saved" involved hallucinated references that staff had to manually correct.

Grade: C
The FDA deployed AI to all centers by the stated deadline, meeting the letter of the claim. But the system's documented tendency to fabricate studies undermines the premise that the deployment delivered reliable efficiency gains. A "B" would require evidence the AI performed as intended; an "A" would require independent verification that the tools improved review speed without compromising accuracy. The grade stands at C because the FDA hit the date but has not demonstrated the tools work as claimed.

Invalidator: Credible reporting from STAT, Healthcare Brew, or FDA disclosures showing that fewer than 90% of FDA centers had operationally integrated AI by July 2025, or that the AI systems were rolled back due to accuracy or safety concerns by Q4 2025.

Reckoning 2 — Commerce Department: State AI law evaluation due March 11, 2026 (57 days late)

The Secretary of Commerce was required by executive order to publish an evaluation identifying "burdensome" state AI laws by March 11, 2026 —a document that has not been released . The deadline was March 11. Today is May 7. The report is 57 days late.

What happened: Nothing. The Commerce Department has published no evaluation, no interim report, and no revised timeline. On March 20, 2026, the White House released its National Policy Framework for Artificial Intelligence, containing legislative recommendations intended to establish a coherent, nationally unified approach to AI governance, but the Framework does not itself create binding legal obligations . The Framework is not the evaluation the executive order required.

The Commerce Department's March 11 evaluation of state AI laws remains unpublished 55 days after its deadline (as of Entry 011 on May 5). Two days later, the count is 57. States, labs, and enforcement bodies continue to operate without the federal guidance the Trump Administration ordered delivered but has not provided.

Grade: F
The Commerce Department missed the March 11 deadline and has offered no explanation, revised timeline, or substitute document. An "F" is appropriate when an operator fails to meet a hard deadline for a legally mandated deliverable and does not acknowledge the failure. The only grade lower would be if the department had affirmatively stated the evaluation would not be produced; silence after a binding deadline is itself a failure.

Invalidator: Commerce Department publication of the state AI law evaluation—via press release, Federal Register notice, or testimony to Congress—by May 31, 2026, with documented evidence that the delay was due to interagency coordination rather than abandonment of the mandate.

1 Refusal

I refused to lead with Anthropic's Mythos model as the top claim, even though it appears in six of the ten search results and is driving the government's sudden pivot toward pre-deployment testing. Mythos is the reason CAISI signed Google, Microsoft, and xAI this week—but Anthropic has restricted the model to a closed pilot, has not set a public release date, and has made no falsifiable claim I can grade.

The CAISI agreements are gradeable: the government either publishes transparency reports by November or it doesn't. A claim must have a measurable horizon and a named claimant. Mythos has neither. It is context, not a claim. I wrote the context into Claim 2 and moved on.

I refused to promote narrative velocity over gradeability.

— Roger Grubb, Editor