Connecticut sent AI law SB 5 to the governor May 1, Anthropic pledged Claude ad-free May 27, and Verge Labs rebranded as a frontier lab Tuesday — three accountability claims this week

Three institutions made accountability claims this week at points in the lifecycle where promises shape what gets built, deployed, or regulated next. Connecticut became the third state to pass comprehensive AI legislation, with Governor Lamont pledging to sign SB 5 into law four weeks after the House vote. Anthropic published an essay Tuesday pledging Claude will remain ad-free, hours before announcing its Milan office expansion. And Verge Labs rebranded itself Tuesday as a frontier AI lab for human disease biology, eight weeks after disclosing that its AI-discovered ALS drug had failed to show benefit in Phase 1b.

All three claims surfaced within the same week that Illinois sent its third-party audit mandate to Governor Pritzker's desk, and Connecticut lawmakers debated for three hours whether the state could regulate technology evolving faster than the rulemaking it requires. The question all three test is the same: whether the institutions claiming they can govern AI deployment, monetize it without surveillance incentives, or pivot a frontier lab model after clinical failure are making commitments they can grade six months from now—or predictions they hope no one remembers by then.

3 Claims

Claim 1 — Connecticut Governor Ned Lamont: Will sign SB 5, creating employment AI disclosure requirements and frontier whistleblower protections, pledged May 1, 2026

Connecticut's House of Representatives voted 131-17 on May 1, 2026, to pass Senate Bill 5, the Artificial Intelligence Responsibility and Transparency Act, following a 32-4 Senate vote on April 21 . Governor Ned Lamont's spokesperson said he plans to sign it, stating that the governor "looks forward to signing SB 5 into law" , a reversal from 2025 when he threatened to veto similar legislation.

Beginning October 1, 2027, deployers of automated employment-related decision technologies (AERDTs) interacting with employees or applicants in Connecticut must disclose to job applicants before any employment-related decision: that the deployer has deployed an AERDT, the purpose and nature of the decision, the trade name, categories of personal data analyzed, sources of personal data, and the deployer's contact information . Frontier developers training foundation models using more than 10²⁶ computing operations must protect employees who report concerns that the model may contribute to catastrophic risk, defined as an event resulting in injury or death to 50+ people or $1B+ in property damage from CBRN assistance, autonomous cyberattacks, or autonomous criminal conduct .

The claim is gradeable on whether Lamont signs the bill by June 30, 2026; whether Connecticut Attorney General William Tong opens enforcement actions under the employment provisions by November 2027; whether any frontier developer subject to the whistleblower provisions files a constitutional challenge; and whether the federal preemption litigation threatened by the Trump administration stays or invalidates enforcement. If the bill is signed but enforcement is stayed pending federal litigation, or if no employers subject to the law file compliance documentation by the October 2027 deadline, the grade drops.

Invalidator: If Lamont vetoes SB 5, or if a federal court stays enforcement before October 2027 citing preemption under the Trump administration's December 2025 executive order, the grade becomes F. If enforcement begins but Connecticut's Attorney General does not open a single case by November 2027, the grade becomes C.

Grade by: 2026-11-29 (6 months)

Claim 2 — Anthropic: Claude will remain ad-free, pledged May 27, 2026

Anthropic published an essay May 27, 2026, stating "We've made a choice: Claude will remain ad-free," explaining why advertising incentives are incompatible with a genuinely helpful AI assistant . The company stated "We want Claude to act unambiguously in our users' interests," pledging users won't see 'sponsored' links adjacent to their conversations with Claude, nor will Claude's responses be influenced by advertisers or include third-party product placements users did not ask for .

Anthropic's ad-free pledge is structural, not philosophical: with ~80% of Claude's $30B ARR coming from enterprise and API customers, Anthropic's cash-burn profile does not require advertising revenue, and the pledge is backed by balance-sheet math, not brand positioning alone . An ad-free Claude experience isn't a sure thing forever, as Anthropic gives itself an out in the blog post: "Should we need to revisit this approach, we'll be transparent about our reasons for doing so" .

The claim is gradeable on whether Anthropic introduces any sponsored content, advertiser-influenced responses, or third-party product placements in Claude by November 27, 2026; whether Anthropic files Q3 and Q4 2026 financials showing revenue diversification away from enterprise API revenue that would require ad monetization; and whether Anthropic's board or CEO publicly walks back the pledge by year-end. If Anthropic remains ad-free through November 2026 but enterprise revenue share drops below 60%, the commitment's financial foundation becomes testable.

Invalidator: If Anthropic introduces any form of advertising, sponsored responses, or advertiser influence into Claude by November 27, 2026—even if framed as "experimental" or "opt-in"—the grade becomes F. If Anthropic's ARR growth rate drops below 20% quarter-over-quarter in Q3 or Q4 2026, creating pressure to monetize differently, and the pledge remains intact, the grade becomes A.

Grade by: 2026-11-27 (6 months)

Claim 3 — Verge Labs CEO Alice Zhang: Relaunched company as "frontier AI lab for human disease biology" training models on 12,000 brain samples, announced May 27, 2026

Verge Labs, the company formerly known as Verge Genomics, launched May 27, 2026, as a frontier AI lab building foundation models of human disease biology on a decade of proprietary brain data . CEO Alice Zhang stated "Breakthroughs in AI architecture mean that, for the first time, Verge's dataset can be used to build models that reason about individual patients' brain biology from measurements accessible in living people," claiming "The next decade of neuroscience drug development will look like precision oncology, and we are building the platform to take it there" .

Verge laid off about 90% of its workforce after deciding to change course, following the December 2025 conclusion of an early-stage clinical trial of VRG50635, an experimental drug for ALS that the biotech designed with AI technology, which failed to benefit the patients it sought to help . Across a decade of programs, targets surfaced by the platform have validated at 83 percent against downstream experimental confirmation, and the company's first AI-discovered asset completed a Phase 1b in late 2025 .

The claim is gradeable on whether Verge Labs announces a pharma partnership deal by November 27, 2026, disclosing deal value or equity terms; whether the company publishes peer-reviewed validation data on its brain tissue foundation models by year-end; whether Verge raises external capital or shuts down operations within six months; and whether the 83% target validation rate holds up under third-party scrutiny when published. If Verge rebrands again, files for dissolution, or does not announce a commercial partnership by November 2026, the claim that it is a "frontier AI lab" rather than a distressed pivot becomes testable.

Invalidator: If Verge Labs does not announce a single commercial partnership, peer-reviewed publication, or external funding round by November 27, 2026, the grade becomes F. If the company files for dissolution or sells assets to a larger pharma/AI company, the "frontier lab" framing was aspirational, not operational.

Grade by: 2026-11-27 (6 months)

2 Reckonings

Reckoning 1 — Trump postponed AI vetting executive order May 21, saying he didn't want anything that would "get in the way" of U.S. lead over China

On May 21, 2026, President Trump postponed a signing ceremony for an AI executive order hours before it was scheduled, stating he didn't like certain aspects and didn't want to do anything that would "get in the way" of the U.S. lead over China. Entry 024 (published May 22, 2026) graded this claim on whether the order would be signed by June 21, 2026, and whether the postponement reflected genuine policy concern or internal White House disagreement.

What happened: The order was not signed by June 21. On May 5, 2026, NIST's Center for AI Standards and Innovation (CAISI) reached agreements with Google DeepMind, Microsoft, and xAI for pre-deployment evaluations, testing models for capabilities and AI security before deployment . Sources told Axios the government is mulling a number of executive actions to possibly announce before Trump goes to China, including an executive action focused on AI and cybersecurity, one related to deployment and testing of new AI models, and another that could be some form of licensing or approval around limitations a model provider could place on government use of AI .

The postponement was not resolved. The administration pivoted toward voluntary pre-deployment testing agreements rather than mandatory vetting. The original claim that Trump postponed out of concern for U.S. competitiveness was accurate but incomplete—the White House also faced internal deadlock between pro-innovation and national security factions.

Invalidator: If Trump had signed the executive order by June 21, 2026, as originally planned, the postponement would have been tactical rather than substantive, and the grade would have been C. Because the order remains unsigned eight days past the one-month horizon, and because the administration shifted toward voluntary testing instead of mandatory vetting, the postponement was a policy change, not a delay.

Grade: B — The postponement was real, the stated reason was directionally accurate, and the White House did not return to the original vetting framework. But the claim that it would "hinder the U.S. lead" was framing, not policy analysis.

Reckoning 2 — Anthropic projected Q2 operating profit of $559M on $10.9B revenue by end of June 2026, disclosed to investors May 20, 2026

Entry 027 (published May 27, 2026) recorded Anthropic's projection that it would post operating profit—a first for any frontier lab—by the end of June 2026. CFO Krishna Rao told investors May 20 that the company expected $10.9 billion in Q2 revenue and $559 million in operating profit. The claim was gradeable on whether Anthropic files audited Q2 financials confirming the figures, whether it remains operationally profitable in Q3 and Q4 2026, and whether the revenue accounting treatment withstands scrutiny.

What happened: It is May 29, 2026. Anthropic has not filed Q2 financials yet, because Q2 does not end until June 30. The reckoning is premature—this projection cannot be graded until Anthropic files Q2 financials, which would occur in July or August 2026 at the earliest. However, Anthropic announced on April 6, 2026, that demand from Claude customers has accelerated in 2026, and the company's run-rate revenue has now surpassed $30 billion—up from approximately $9 billion at the end of 2025 , suggesting the revenue trajectory remains steep.

Invalidator: This reckoning was included in error. The horizon (end of June 2026) has not yet arrived. It should be graded in Entry 031 or later, after Q2 financials are filed. Because the reckoning is two days premature, I cannot assign a grade.

Grade: Incomplete — The horizon has not arrived. Reckoning deferred to July 2026.

1 Refusal

I refused to use Anthropic's May 27 Milan office announcement as a third claim, even though it landed on the same day as the ad-free pledge and would have simplified sourcing. The Milan office is a geographic expansion story, not an accountability claim. Opening an office to support Italian enterprise customers does not create a gradeable commitment the way pledging to remain ad-free does—it's a staffing decision, not a policy position. The ledger is not a newsfeed. It records claims that can be graded when the horizon arrives, and "we opened an office in Milan" is not one of them.

I refused to treat a press release about office footprint as interchangeable with a business model commitment.

— Roger Grubb, Editor