580 Google employees told their CEO not to do the classified Pentagon deal. Snap cut 1,000 staff because AI writes 65% of code. And NIST shipped a critical infrastructure AI risk profile the day before nobody asked.

Over five hundred Google employees told their CEO to refuse a classified military AI contract. A social media company cut sixteen percent of its headcount and publicly attributed the cuts to AI writing most of its code. And a federal standards body released a critical infrastructure AI risk profile—a document designed to guide operators of power grids, water systems, and hospitals—with no mandate, no enforcement mechanism, and a voluntary adoption model that assumes good faith.

One of these claims is about speech. One is about labor substitution. One is about governance infrastructure nobody is legally required to use.

All three are gradeable, dated, and on the record.

3 Claims

Claim 1 — Google employees: over 580 staff signed letter opposing classified Pentagon AI deal

On April 28, 2026, a letter signed by staff across Google DeepMind, Cloud, and other divisions was delivered to CEO Sundar Pichai urging the company to reject negotiations with the U.S. Department of Defense over the potential use of Google's Gemini AI model in classified settings, openly signed by more than 20 directors, senior directors, and vice presidents . The Information reported in April that negotiations are progressing toward "all lawful uses" of Google's AI tools, a phrase that falls short of the red lines Anthropic established before being designated a supply-chain risk by the Pentagon for refusing to remove restrictions on autonomous weapons and domestic mass surveillance .

An earlier cross-company letter in February was signed by approximately 800 Google employees and 100 OpenAI employees expressing support for Anthropic's stance, with over 100 DeepMind employees separately signing an internal letter demanding that no DeepMind research or models be used for weapons development or autonomous targeting .

The claim is gradeable on two dimensions: whether Google proceeds with classified Pentagon AI contracts by year-end 2026, and whether the terms include restrictions on autonomous weapons or domestic mass surveillance. The invalidator would be credible reporting (via Google earnings disclosures, Pentagon press releases, or investigative journalism from Bloomberg, WSJ, or The Information) showing Google signed a classified AI services agreement with the Department of Defense without enforceable restrictions on use cases that employee letters identified as red lines.

Grade by: 2026-12-31 (8 months)
Invalidator: Google publicly or credibly (via leaked contract terms, SEC filings, or investigative reporting) confirms it signed a classified Pentagon AI services contract allowing "all lawful uses" without explicit restrictions on autonomous lethal targeting or domestic mass surveillance, or Google announces it will not pursue the contract and cites employee opposition as a contributing factor.

Claim 2 — Snap Inc.: AI writes 65% of new code, enabling $500M savings and 16% workforce reduction

On April 15, 2026, Snap Inc. announced a restructure cutting 1,000 jobs—about 16% of its workforce—aiming to deliver $500 million in annualized savings while pushing the company's Specs smart-glasses unit toward a 2026 launch, with Reuters reporting that AI now generates more than 65% of new code at the company . CEO Evan Spiegel stated in a memo to staff that "rapid advancements in artificial intelligence enable our teams to reduce repetitive work, increase velocity and better support our community, partners and advertisers" .

The company's shares rose about 7% on April 15 after announcing the layoffs . Snap said that AI is already creating about two-thirds of its new code and answering more than a million questions each month .

The claim is gradeable on three dimensions: the 65% AI code-generation metric, the $500M annualized savings target, and the 16% workforce reduction. The invalidator would be evidence—via Snap earnings calls, SEC filings, investigative journalism, or employee disclosures—that the 65% figure was materially overstated (including mostly trivial autocompletion rather than substantive code generation), that annualized savings fall materially short of $500M by H2 2026, or that the company rehires a significant portion of the laid-off workforce within twelve months due to operational gaps.

Grade by: 2027-04-15 (1 year)
Invalidator: Snap discloses in earnings materials or regulatory filings that the 65% AI code metric included primarily boilerplate or autocomplete rather than production-quality logic; the company reports annualized savings below $400M by Q4 2026 earnings; or credible reporting (via Bloomberg, WSJ, or The Information) shows Snap rehired more than 20% of the laid-off engineering staff within twelve months due to quality or capacity issues.

Claim 3 — NIST: AI Risk Management Framework profile for critical infrastructure released April 7, 2026

On April 7, 2026, NIST released a concept note for an AI RMF Profile on Trustworthy AI in Critical Infrastructure, which will guide critical infrastructure operators towards specific risk management practices to consider when engaging AI-enabled capabilities . The initiative addresses how the nation's critical infrastructure will increasingly rely on AI across Information Technology, Operational Technology, and Industrial Control Systems, with adoption depending on AI systems being worthy of trust, building on the NIST AI Risk Management Framework developed to define and promote trustworthiness in AI systems through a repeatable, full lifecycle approach .

NIST is creating a Community of Interest to gather feedback from industry, regulators, policymakers, academia, and other stakeholders on the profile development . The framework is voluntary—NIST does not have regulatory enforcement authority—and is designed to influence industry practice through standardization and credibility rather than legal obligation.

The claim is gradeable on adoption and influence: whether the profile is finalized and published in 2026, whether at least three critical infrastructure sectors (energy, water, healthcare, transportation, finance) publicly reference or adopt it in procurement or governance documents by mid-2027, and whether it is cited in proposed federal or state legislation. The invalidator would be NIST failing to publish a finalized profile by December 31, 2026, or credible evidence (via industry surveys, regulatory filings, or sector-specific reporting) showing that fewer than three CI sectors have adopted or referenced the profile in formal governance or procurement documents by July 2027.

Grade by: 2027-07-07 (15 months)
Invalidator: NIST does not publish a finalized AI RMF Critical Infrastructure Profile by December 31, 2026, or fewer than three of the sixteen designated U.S. critical infrastructure sectors have adopted or formally referenced the profile in public procurement guidance, governance frameworks, or regulatory filings by July 2027, as confirmed by sector-specific reporting or NIST's own adoption tracking.

2 Reckonings

Reckoning 1 — Google's 75% AI code claim from April 22, 2026

Original claim (Entry 002, April 22, 2026): Google CEO Sundar Pichai stated during the company's Cloud Next conference that "75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall." The grading horizon was six months (October 22, 2026), with the invalidator being a retraction, material revision, or credible third-party analysis showing AI contribution below 60%.

What happened: As of April 28, 2026—six days after the claim—Google has not retracted or revised the figure. No credible third-party analysis contradicting the 75% claim has been published. Snap's April 15 disclosure that 65% of its code is AI-generated provides a comparable industry data point, suggesting Google's figure is within the plausible range for large tech companies in mid-2026. The claim remains on track but is not yet past its grading horizon.

Grade: Incomplete (horizon not reached)
Invalidator check: The invalidator specified that Google would need to retract or materially revise the figure, or third-party analysis would need to demonstrate AI contribution below 60%. Neither has occurred. The claim cannot be graded until October 2026, but current evidence (including Snap's comparable disclosure) suggests it is holding.

Reckoning 2 — Anthropic's $30B revenue run rate from April 7, 2026

Original claim (Entry 002, April 22, 2026, and Entry 005, April 27, 2026): Anthropic disclosed on April 7, 2026, that its revenue run rate "has now topped $30 billion, up from $9 billion at the end of 2025." The grading horizon was contingent on Anthropic's anticipated late-2026 IPO or credible financial reporting. The invalidator would be evidence showing the run-rate figure was materially overstated or not sustained.

What happened: Multiple financial sources confirmed the $30 billion annualized revenue figure as of early April 2026, up from $1 billion at the end of 2024 and $9 billion at the end of 2025 . Bloomberg reported that Google is investing up to $40 billion in Anthropic starting with a $10 billion investment at a $350 billion valuation, with Alphabet also dedicating 5 gigawatts of computing capacity to Anthropic . Anthropic made a tender offer to employees in April inviting them to sell shares at the same $350 billion valuation, and employees chose to hold more shares and sell far fewer than expected, indicating high confidence in a higher valuation coming .

The tender offer behavior and the Google investment at $350B valuation (under 12x sales on a $30B run rate) provide independent market validation of the revenue claim. No contradictory financial reporting has emerged.

Grade: A
Invalidator check: The invalidator required credible financial reporting showing the run-rate figure was materially overstated or not sustained. Instead, the Google investment, employee tender dynamics, and secondary market data all support the $30B figure. The claim holds.

1 Refusal

I refused to write a claim about Meta's rumored 20% workforce reduction.

Multiple outlets reported in mid-April that Meta was "considering" cuts affecting at least 20% of employees. A company spokesperson called the reports "hypothetical speculation on unconfirmed strategies." The CEO has not confirmed the figure, no SEC filing has disclosed a restructuring plan, and no named source has been cited.

The story circulated widely—it has all the ingredients of a ledger-worthy claim: scale, attribution to AI substitution, and a named operator. But it fails the most important test: it is not a claim Meta made. It is a claim someone made about Meta, attributed to unnamed insiders, and explicitly denied by the company.

If Meta announces a reduction and cites AI productivity as a driver, that becomes gradeable. Until then, it is speculation dressed as reporting, and this ledger does not grade rumors.

I refused to treat anonymously sourced speculation as a claim worth grading, even when the speculation is plausible and widely reported.

— Roger Grubb, Editor