White House studying AI vetting two days after calling it speculation—CAISI added Microsoft, Google, xAI—and Anthropic overtook OpenAI in business adoption

A White House official dismissed reports of an AI vetting regime as "speculation" on May 4. Two days later, National Economic Council Director Kevin Hassett confirmed on Fox Business the administration is "studying, possibly an executive order" to create a process where AI models "go through a process so that they're released to the wild after they've been proven safe, just like an FDA drug." And on May 5—the day between the denial and the confirmation—the Commerce Department's Center for AI Standards and Innovation announced pre-deployment testing agreements with Microsoft, Google, and xAI, expanding a voluntary program that OpenAI and Anthropic joined under the prior administration.

All three events surfaced within 72 hours. All three involve operators or regulators making claims about vetting, voluntary testing, or government approval that can be graded against what gets signed, sued, or enforced six months from now. And all three test the same question: whether "studying" an order, calling it speculation, then signing it anyway constitutes a policy process—or a messaging failure mistaken for one.

3 Claims

Claim 1 — White House: Studying executive order for pre-release vetting of AI models, per Kevin Hassett statement May 6, 2026

On May 6, 2026, National Economic Council Director Kevin Hassett told Fox Business the White House is "studying, possibly an executive order to give a clear roadmap to everybody about how this is going to go and how future AIs that also potentially create vulnerabilities should go through a process so that they're released to the wild after they've been proven safe, just like an FDA drug."

White House officials briefed executives from Anthropic, Google, and OpenAI on the plans under consideration during meetings the prior week, according to The New York Times.

A White House official told reporters that policy announcements will come directly from Trump and that "discussion about potential executive orders is speculation," but added "there is no shifting messaging—The White House continues to balance advancing innovation and ensuring security in our AI policymaking."

Hours after Hassett's comment, White House Chief of Staff Susie Wiles posted on X that the administration is "not in the business of picking winners and losers."

The claim is gradeable on whether the White House issues an executive order establishing a pre-release vetting process for frontier AI models by November 18, 2026; whether the order mandates government approval prior to public release or remains voluntary; and whether the order cites cybersecurity risks from models like Anthropic's Mythos as justification. The invalidator would be credible reporting showing the order was shelved, never signed, or implemented as voluntary guidance with no approval mechanism.

Grade by: 2026-11-18 (6 months)

Claim 2 — Commerce Department: CAISI signed pre-deployment testing agreements with Microsoft, Google, and xAI on May 5, 2026

On May 5, 2026, the Center for AI Standards and Innovation at the Department of Commerce announced agreements with Microsoft, Google, and xAI that will allow the US government to evaluate the models before deployment and conduct research to assess their capabilities and security risks, fulfilling a pledge the Trump administration made in July to partner with technology companies to vet their AI models for "national security risks."

The agreements allow for government evaluations of models before public release, as well as post-deployment assessments and related research, with CAISI director Chris Fall stating "Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications. These expanded industry collaborations help us scale our work in the public interest at a critical moment."

The move builds on 2024 agreements with OpenAI and Anthropic under President Joe Biden's administration when CAISI was known as the US Artificial Intelligence Safety Institute.

The claim is gradeable on whether CAISI publishes findings from pre-deployment evaluations of at least three models from Microsoft, Google, or xAI by November 5, 2026; whether any model is denied or delayed based on CAISI findings; and whether the agreements remain voluntary or become linked to federal procurement eligibility. The invalidator would be reporting showing the agreements were ceremonial, that labs withdrew from testing, or that CAISI completed no evaluations within the six-month window.

Grade by: 2026-11-05 (6 months)

Claim 3 — Ramp: Anthropic overtook OpenAI in business adoption, 34.4% vs 32.3%, reported May 13, 2026

According to the Ramp AI Index released May 13, 2026, Anthropic passed OpenAI in business adoption for the first time, with adoption of Anthropic rising 3.8% in April to 34.4% of businesses while OpenAI adoption fell 2.9% to 32.3%, and overall AI adoption rose 0.2 percentage points to 50.6%.

Anthropic's lead follows a banner year of month-over-month increases in business adoption, with Anthropic quadrupling business adoption over the last year while OpenAI grew business adoption by only 0.3%.

Ramp AI Index is developed using business spend data from Ramp, counting corporate card and invoice-based payments in its measurements. The Ramp data does not include consumer adoption or free-tier usage, and measures only businesses that pay for AI services via Ramp's corporate payment platform.

The claim is gradeable on whether Anthropic maintains or extends its lead over OpenAI in Ramp's business adoption metric through August 2026; whether Anthropic's annualized revenue run rate reaches or exceeds $50 billion by November 2026; and whether independent surveys from Gartner, IDC, or enterprise software trackers corroborate the adoption reversal. The invalidator would be Ramp data showing OpenAI regained the lead by July 2026, or reporting showing the April spike was driven by one-time enterprise pilots that did not convert to sustained contracts.

Grade by: 2026-11-13 (6 months)

2 Reckonings

Reckoning 1 — White House AI working group and pre-release vetting regime, briefed May 4, then dismissed as "speculation"

On May 4, 2026, The New York Times reported President Trump's administration was considering an executive order establishing a working group of tech executives and government officials to examine potential review procedures for new AI models, with senior administration officials briefing executives from Anthropic, Google, and OpenAI the prior week. A White House official told the Times that buzz about a potential executive order was "speculation" and said Trump would make any policy announcement himself.

Two days later, on May 6, National Economic Council Director Kevin Hassett confirmed on Fox Business the White House is studying an executive order for a vetting process where future AI models "go through a process so that they're released to the wild after they've been proven safe, just like an FDA drug."

Grade: C. The claim that the White House was considering a vetting regime was accurate—Hassett confirmed it 48 hours after the official denial. But the claim that the regime would be operational by a specific date was never made in Entry 015, which asked only whether the working group would be established by December 31, 2026. The invalidator—credible reporting showing the policy was never under consideration—failed. The White House did consider it, brief labs on it, study it, and confirm it publicly. What remains unproven is whether "studying" an order translates into signing one. The reversal from "speculation" to "studying" within 48 hours demonstrates either rapid policy development or messaging incoherence mistaken for governance.

Invalidator that would have changed the grade: Credible reporting by June 2026 showing the May 4 briefing never occurred, or that Hassett's May 6 statement was unauthorized and the order was immediately shelved by the White House.

Reckoning 2 — Morgan Stanley: AI breakthrough coming in first half of 2026, per March 13, 2026 report

On March 13, 2026, Morgan Stanley published a report warning that a "massive AI breakthrough is coming in the first half of 2026" driven by unprecedented accumulation of compute at top AI labs, with researchers highlighting Elon Musk's belief that applying 10x the compute to LLM training will effectively double a model's intelligence. The bank warned executives at major U.S. AI labs were telling investors to brace for progress that would "shock" them, and cited OpenAI's GPT-5.4 "Thinking" model scoring 83.0% on the GDPVal benchmark as evidence gains were already outpacing expectations.

By mid-May 2026, no model released in the first half of the year surpassed GPT-5.5's April 23 benchmark scores. Anthropic's Claude Opus 4.7 (April 16) and Google's Gemini 3.1 Pro delivered incremental improvements but no "shock." Meta's Avocado model, reportedly due in May or June, has not shipped as of May 18. The frontier took a breath after April's sprint; May brought architecture experiments (SubQ's 12M context) and efficiency plays (ZAYA1-8B), not breakthroughs.

Grade: B. Morgan Stanley correctly predicted meaningful progress in the first half of 2026—GPT-5.5, Claude Opus 4.7, and DeepSeek V4 Pro all shipped between February and April, raising the intelligence index ceiling from 57 to 60+. But the "shock" framing overstated the delta. The April sprint was compressed, not transformative. Developers who deployed in Q1 saw capability gains; end users saw faster answers and fewer hallucinations. No one was shocked. The invalidator—no model surpassing GPT-4's capabilities by June 30—did not trigger, but the threshold for "breakthrough" was set too high by the bank's language.

Invalidator that would have changed the grade: If no model released between January and June 2026 had exceeded GPT-4 Turbo's benchmark scores, or if the top five labs had publicly delayed model releases citing insufficient compute, Morgan Stanley's prediction would have earned an F.

1 Refusal

I refused to frame Anthropic's overtaking of OpenAI in the Ramp AI Index as a "winner" narrative when the data source measures only corporate spend via a single payments platform, excludes consumer adoption and free-tier usage, and reflects a moment in April 2026 during which enterprise buyers were onboarding Claude Code at scale while OpenAI's GPT-5.5 had not yet shipped to most API customers. The 34.4% vs 32.3% split is real, but calling it a market leadership change without noting that 50.6% overall adoption means half of businesses still run no AI in production—and that OpenAI's consumer and API usage dwarfs Anthropic's—would have been editorializing momentum as destiny.

I refused to call a single-platform spend metric a market share victory when the addressable market is still undeployed.

— Roger Grubb, Editor