Anthropic told Congress Alibaba ran 28.8 million Claude exchanges, NYC required bias review for 1.1 million students, and Google's June model missed June—three claims this week

One company told Congress a rival ran nearly 29 million API calls to train a competing model. One city school district mandated that every AI tool serving 1.1 million students must pass bias and equity vetting before deployment. And one search giant promised a flagship model by the end of June—then missed the deadline it set in front of 10,000 developers. Three accountability claims landed within sixteen days, and each involves an operator or regulator making an on-the-record statement about industrial-scale enforcement, educational governance, or release timelines that can be graded against what federal prosecutors actually pursue, what vendors actually demonstrate, and what Google actually ships six to twelve months from now.

3 Claims

Claim 1 — Anthropic: Sent June 10, 2026, letter to Senate Banking Committee alleging Alibaba-affiliated operators ran 28.8 million Claude exchanges via 25,000 fraudulent accounts

Anthropic sent a letter dated June 10, 2026, to Senate Banking Committee Chair Tim Scott and Ranking Member Elizabeth Warren alleging that operators affiliated with Alibaba and its Qwen AI lab carried out 28.8 million exchanges with Claude using roughly 25,000 fraudulent accounts between April 22 and June 5 . The letter claims Alibaba carried out "the largest known distillation attack on Anthropic to date."

Anthropic accused Alibaba of conducting the largest adversarial distillation campaign ever documented against its Claude AI, marking the largest such allegation Anthropic has ever filed and the first time a company of Alibaba's global commercial scale has been named in a distillation complaint . Senators Bill Hagerty (R-TN) and Andy Kim (D-NJ) are moving to introduce an amendment to defense legislation that would blacklist or sanction entities found conducting adversarial distillation campaigns .

The claim is gradeable: either the Justice Department opens an investigation, Congress passes legislation imposing penalties on distillation operators, or enforcement remains limited to terms-of-service violations by December 2026.

Claimants: Anthropic; U.S. Senate Banking Committee (recipients)
Grade by: 2026-12-26 (6 months)
What would invalidate the claim: No federal investigation opened, no sanctions legislation passed, and no enforcement action taken beyond civil contract disputes by December 26, 2026, demonstrating that naming Alibaba produced no material federal response.

Claim 2 — New York City Department of Education: Issued preliminary guidance requiring bias and equity review for all AI tools deployed across 1.1 million students, with comprehensive playbook scheduled for June 2026

NYC's Department of Education mandates bias and equity vetting for all AI tools across its 1.1 million-student system, with a complete policy framework scheduled for release by June 2026, requiring no AI tool to enter classrooms without first clearing a bias and equity review . Every AI tool must pass a vetting process before school-level deployment, including evaluation for algorithmic bias, equity impact, cultural responsiveness, and developmental appropriateness, with a comprehensive playbook planned for June 2026 .

NYCPS is actively building the capacity to review for algorithmic bias, equity impact, and instructional effectiveness, with this expanded review to be reflected in the upcoming AI Playbook, and topics under active development to be included in the June 2026 Playbook . A full AI policy playbook is scheduled for release in June 2026, and for companies already embedded in NYC schools, the timeline is immediate .

The claim is gradeable: NYC DOE either publishes a binding compliance playbook with enforceable bias and equity standards for AI tools by June 30, 2026, or the preliminary guidance remains aspirational without teeth.

Claimant: New York City Department of Education
Grade by: 2026-06-30 (4 days)
What would invalidate the claim: No comprehensive playbook published by June 30, 2026, or playbook published without binding algorithmic bias evaluation requirements, demonstrating the June deadline was not met or the vetting framework remains voluntary guidance rather than enforceable policy.

Claim 3 — Google: CEO Sundar Pichai stated May 19, 2026, at Google I/O that Gemini 3.5 Pro would roll out "next month" (June 2026), setting public expectation for general availability

Google announced Gemini 3.5 Pro at I/O 2026 alongside Gemini 3.5 Flash on May 19, with Sundar Pichai's exact words on stage: "We're also hard at work on 3.5 Pro. It's already being used internally, and we look forward to rolling it out next month," meaning June 2026 . Google has postponed the release of Gemini 3.5 Pro to July 2026, according to insider reports, with the model remaining in internal testing and limited enterprise preview while its sibling model Gemini 3.5 Flash is already available .

Gemini 3.5 Pro is still not out, slipping past its planned June 2026 general release as the month nears its end, with Google unveiling it at I/O on May 19, 2026, and Sundar Pichai telling the I/O audience to wait roughly another month for broad access . With 11 days remaining in June, the window is closing, and Polymarket prediction markets give roughly 50–55% odds for a pre-June 30 release .

The claim is gradeable: either Google ships Gemini 3.5 Pro to general availability (not just limited enterprise preview) by June 30, 2026, or the company missed the timeline the CEO set on stage in front of 10,000 developers.

Claimant: Google / Sundar Pichai
Grade by: 2026-06-30 (4 days)
What would invalidate the claim: Gemini 3.5 Pro does not reach general availability (public API, Google AI Studio, or consumer Gemini app access) by June 30, 2026, demonstrating that Pichai's "next month" commitment was not honored and the model slipped into July or later.

2 Reckonings

Reckoning 1 — Trump June 2 executive order: Voluntary 30-day pre-release testing framework for frontier AI models

On June 2, 2026, President Trump signed an executive order establishing a voluntary framework asking AI companies to submit their most powerful models for government testing up to 30 days before releasing them to the public. Entry 047 graded this claim with a six-month horizon ending December 2, 2026. The claim was that frontier AI developers would voluntarily submit models for pre-release government review.

What has happened: As of June 26, 2026, no major lab has publicly confirmed voluntary submission of a model for pre-release government testing under the framework. Anthropic released Fable 5 and Mythos 5 on June 9 without prior voluntary submission; the government ordered suspension on June 12 via export control directive—an involuntary action, not a voluntary pre-release review. OpenAI has not announced submission of GPT-5.6 for voluntary testing. Google has not submitted Gemini 3.5 Pro.

The voluntary framework has generated zero publicly confirmed pre-release submissions in the first 24 days. What did occur was involuntary export control enforcement—the opposite of voluntary cooperation.

Grade: C
The executive order created the legal structure for voluntary testing, but no evidence exists that any frontier lab voluntarily submitted a model for 30-day pre-release review rather than proceeding directly to launch. The government achieved model suspension through export controls (mandatory), not through the voluntary framework (unused).

Invalidator: If any frontier lab—OpenAI, Anthropic, Google, or Microsoft—had publicly announced voluntary submission of a model for pre-release government testing by June 26, the grade would have been higher. The absence of voluntary participation, combined with the government resorting to mandatory export controls for Anthropic's models, demonstrates the framework has not achieved its stated purpose.

Reckoning 2 — EU AI Act high-risk enforcement: August 2, 2026, deadline for Annex III high-risk AI systems

Entry 048 (June 25, 2026) reported that the EU Parliament voted June 16 to delay high-risk AI obligations until December 2, 2027, postponing the August 2, 2026, enforcement deadline by sixteen months. The original claim—that high-risk obligations would become enforceable August 2, 2026—was graded with a six-month horizon in Entry 045 (June 22, 2026).

What has happened: The EU Parliament voted June 16, 2026, to approve the Omnibus amendment postponing stand-alone Annex III high-risk system obligations from August 2, 2026, to December 2, 2027. The provisional agreement reached May 7, 2026, became final legislation on June 16, eliminating the August 2 deadline before it arrived.

The August 2, 2026, enforcement deadline no longer exists in EU law. Operators of stand-alone high-risk AI systems subject to Annex III classification received a sixteen-month reprieve.

Grade: A (for the reversal claim)
The enforcement deadline that dominated AI governance discussions in Q1 2026 was legislatively eliminated before it took effect. The EU delivered exactly what the May 7 provisional agreement promised: postponement of the high-risk deadline to December 2027. Operators who prepared for August 2 compliance now face December 2027 instead.

Invalidator: If the EU Parliament had rejected the Omnibus amendment on June 16, or if member states failed to ratify the postponement, the August 2, 2026, deadline would have remained in force and the grade would be F. The legislative reversal was completed as promised.

1 Refusal

I refused to treat Google's Gemini 3.5 Pro delay as a "slip" or "postponement" without grading the original claim against the CEO's public commitment. Sundar Pichai told a room of 10,000 developers at Google I/O on May 19 to expect the model "next month." Next month is June. We are four days from June 30, and the model is not in public hands. That is not "reportedly delayed"—it is a missed deadline, gradeable and on the record. I could have written the claim as "Google is refining Gemini 3.5 Pro ahead of an expected July launch," which would have let the company off the hook for a promise the CEO made on stage. I refused to do that.

I refused to soften a public claim into corporate hedging when the speaker named the month and the month is ending.

— Roger Grubb, Editor