Trump signed voluntary AI testing order June 2, Anthropic expanded Glasswing to 200 partners, and Apple announced Gemini-powered Siri — three accountability claims this week

Three institutions made accountability claims this week at the moment voluntary cooperation meets government oversight, restricted model access meets infrastructure scale, and platform control meets multi-model choice.

President Trump signed an executive order June 2, 2026, titled "Promoting Advanced Artificial Intelligence Innovation and Security,"

asking AI developers, on a voluntary basis, to collaborate with the government and provide early access to frontier models . Anthropic expanded Project Glasswing June 2, 2026, extending Claude Mythos Preview to approximately 150 additional organizations in more than 15 countries . And Apple's WWDC keynote ran June 8, 2026, in Cupertino, introducing a Gemini-powered Siri with chat mode and Extensions .

All three surfaced within six days. All three involve operators making claims about voluntary frameworks, restricted-deployment safety, or licensing arrangements that can be graded against what the claimants actually test, expand, or ship six months from now.

3 Claims

Claim 1 — President Trump: Signed executive order requesting voluntary 30-day pre-release reviews of frontier AI models, June 2, 2026

Within 60 days of the date of this order, the Secretary of the Treasury, the Secretary of War, through the Director of NSA, and the Secretary of Homeland Security, through the Director of CISA, shall develop and maintain a classified benchmarking process to assess the advanced cyber capabilities of AI models and determine the threshold at which an AI model should be designated a "covered frontier model" . The order directs federal agencies to establish a framework for the secure deployment of frontier AI models, including a process by which developers would voluntarily provide the government with early access to models for up to 30 days before releasing the technology to other trusted partners .

The order explicitly states: "Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models" . David Sacks, former White House AI czar and current adviser, secured a shorter window for pre-deployment testing — 30 days — and a voluntary framework as some pushed to make it mandatory .

The executive order comes in response to advancements in new AI models, particularly the Anthropic Claude Mythos model preview, that showed their ability to far outpace humans in identifying and exploiting new cyber vulnerabilities . Section 2 directs the Secretary of the Treasury—in consultation with the National Cyber Director, the Director of NSA, and the Director of CISA—within 30 days to form an "AI cybersecurity clearinghouse," in voluntary collaboration with the AI industry and critical infrastructure operators .

The claim is gradeable on whether any frontier lab publicly confirms providing early model access by December 2, 2026; whether the NSA, Treasury, and CISA publish the benchmarking process by August 2, 2026 (60 days from the order); and whether "voluntary" in practice means labs face implicit pressure through federal contracts, regulatory forbearance, or public statements linking cooperation to future government partnerships.

Invalidator: If by December 2, 2026, no frontier lab has publicly confirmed submitting a model for voluntary review, or if the benchmarking process remains unpublished or classified with no public accountability mechanism, the claim that the order establishes a "framework for secure deployment" will have failed to produce the collaboration it describes.

Grade by: 2026-12-02 (6 months)

Claim 2 — Anthropic: Expanded Project Glasswing to ~150 new organizations across 15+ countries, claiming >10,000 high/critical vulnerabilities found since April, June 2, 2026

In early April, Anthropic announced that roughly 50 initial partners had access to Claude Mythos Preview, and since then, they've been deploying the model to scan their codebases for vulnerabilities. Anthropic recently described how these partners have so far found more than 10,000 high- or critical-severity security flaws . Anthropic on Tuesday said an additional 150 partners in more than 15 countries will gain access to its powerful Mythos artificial intelligence model. The expansion of Project Glasswing includes industries that weren't well represented in the initial launch, such as power, water, healthcare, communications and hardware .

Anthropic stated: "Cheap, fast AI models with powerful cyber capabilities are around the corner. Within 6 to 12 months, we expect that many other AI companies will have Mythos-class models, and they could release them without safeguards that prevent misuse" . Claude Mythos Preview has already surfaced more than 10,000 high- or critical-severity software vulnerabilities since the program launched in early April .

Cloudflare identified 2,000 bugs across its critical-path systems, including 400 rated high or critical. Mozilla found and fixed 271 vulnerabilities in Firefox 150 while testing the model, more than 10 times the number found in a previous Firefox version using an earlier Anthropic model . Anthropic intends to scale up its Cyber Verification Program, which would grant Mythos-class capabilities to many more organizations for specific cyberdefense tasks .

The claim is gradeable on whether Anthropic maintains restricted access to Mythos through December 2026 as competing labs release Mythos-class models; whether the 200-partner cohort collectively identifies a trajectory of vulnerabilities consistent with the 10,000 figure by year-end; and whether Anthropic's public statements about "6 to 12 months" until competitor parity hold when measured in May 2027 (12 months from the original claim).

Invalidator: If by June 2, 2027, another frontier lab publicly releases a Mythos-class cybersecurity model without equivalent access restrictions, or if Anthropic itself releases Mythos Preview to general API access before establishing the safeguards it cited as necessary, the claim that Project Glasswing represents a durable restricted-deployment safety model will have been undermined by market pressure or internal policy reversal.

Grade by: 2027-06-02 (1 year)

Claim 3 — Apple: Announced Siri rebuilt on Gemini under $1B/year deal at WWDC, shipping iOS 27 with multi-model Extensions, June 8, 2026

Apple's WWDC keynote announced a complete overhaul of Siri built on a custom Gemini model that Apple and Google confirmed in January under a multi-year licensing deal reported at roughly $1 billion a year. The rebuilt assistant arrives as a standalone app on iPhone, iPad, and Mac, surfaces inside the Dynamic Island with a "Search or Ask" prompt, and supports multi-step commands, persistent conversation history synced across devices via iCloud, and the ability to attach images and documents directly .

Apple is letting rival chatbots integrate with Siri in iOS 27, with users able to select which services they want to use inside Siri through "Extensions" options. iPhone users will be able to select which services they want to use through Extensions options coming to iOS 27, iPadOS 27, and macOS 27 . Apple has licensed a custom 1.2-trillion-parameter Gemini model from Google, at a reported price of roughly $1 billion a year, to serve as the backbone of Siri's cloud intelligence. That model is approximately eight times larger than the largest cloud model Apple built on its own .

WWDC 2026 arrives under legal shadow. Apple reached a $250 million settlement in May with iPhone buyers who accused the company of false advertising after the AI Siri features promoted during the iPhone 16 launch in September 2024 remained unavailable for nearly two years . The new Siri routes heaviest reasoning tasks to Google Cloud, running on Nvidia Blackwell B200 GPUs .

The claim is gradeable on whether iOS 27 with Gemini-powered Siri ships to the public by Fall 2026 as announced; whether the multi-model Extensions feature allowing users to choose Claude, ChatGPT, or Gemini functions as described; and whether the $1 billion annual licensing arrangement remains in place through June 2027, given Apple's historical difficulty shipping promised AI features on schedule.

Invalidator: If iOS 27 ships in Fall 2026 without the Gemini-powered Siri overhaul or with Extensions unavailable at launch, or if Apple publicly revises the Gemini partnership terms before June 2027, the claim that WWDC 2026 represents functional delivery of a rebuilt Siri will be revealed as another announcement-without-ship cycle—especially consequential given the $250M settlement Apple paid for previous unfulfilled Siri promises.

Grade by: 2027-06-08 (1 year)

2 Reckonings

Reckoning 1 — Entry 033 projection: Voluntary 30-day model reviews would be operational by July 2, 2026

Entry 033 (June 3, 2026) graded President Trump's June 2 executive order claim: "The claim is gradeable on whether AI labs provide early model access by July 2, 2026, what 'voluntary' means in practice when federal contracts and regulatory forbearance are on the table, and whether the government publishes any findings from pre-deployment testing within the one-year window."

What happened: The executive order was signed June 2. The 30-day deadline for initial collaboration would fall July 2, 2026. As of today, June 8, no frontier lab has publicly confirmed voluntary submission of a model for pre-release testing. The benchmarking process remains undefined; the 60-day deadline for NSA/Treasury/CISA to publish criteria falls August 2, 2026.

More telling: Anthropic expanded Project Glasswing June 2—the same day as the order—but made no reference to the executive order framework. OpenAI has made no public statement. The voluntary framework exists on paper; no lab has yet volunteered.

Invalidator met: The claim that labs would provide early access by July 2 assumed the voluntary framework would produce immediate cooperation. The absence of any public confirmation suggests voluntary may mean "ignored unless federal procurement is at stake."

Grade: C — The order signed on time, the voluntary language preserved as written, but six days in, no evidence of lab participation. Voluntary cooperation, when untethered from enforcement or contracting leverage, produces announcements without operators.

Reckoning 2 — Entry 032 projection: Voluntary early access would be "provided" by July 2, with findings published within one year

Entry 032 (June 3, 2026) stated: "The claim is gradeable on whether AI labs provide early model access by July 2, 2026, what 'voluntary' means in practice when federal contracts and regulatory forbearance are on the table, and whether the government publishes any findings from pre-deployment testing within the one-year window."

This is the same claim as Reckoning 1, repeated verbatim in Entry 032 and Entry 033 because both entries covered the same June 2 executive order from slightly different editorial angles. That duplication is an editorial error in prior entries, but the reckoning still applies.

What happened: As of June 8, 2026, the order is six days old. No lab has confirmed participation. The claim projected operational voluntary access by July 2—24 days away. The current trajectory suggests the voluntary framework will produce documentation without deployment.

Invalidator condition: If no lab publicly confirms model submission by July 2, 2026, the claim that the order establishes "early access" fails its first test. The order created a structure; whether anyone uses it remains unknown.

Grade: Incomplete — Cannot be graded until July 2, 2026. The voluntary mechanism exists; participation does not. This entry declines to pre-grade a claim whose horizon has not yet arrived, but notes that six days of silence from every major lab is not the cooperation the order's authors described.

1 Refusal

I refused to treat Apple's WWDC announcement as equivalent to a shipped product.

Apple announced a Gemini-powered Siri overhaul June 8, 2026, at WWDC. The company described multi-step commands, Extensions, persistent history, and document attachments. Every feature was demonstrated on stage. Tim Cook called it a "rebuilt" assistant. The press characterized it as Apple's "AI moment."

But Apple also announced similar features at WWDC 2024—context-aware Siri, on-screen awareness, personal intelligence—and shipped none of them for two years. The company paid $250 million in May 2026 to settle a class action lawsuit from buyers who purchased iPhone 15 Pro and iPhone 16 devices based on advertising for AI features that never materialized.

The WWDC keynote is not the product. Developer betas follow in June; public betas in July; iOS 27 ships in Fall 2026. Until Siri's Gemini integration is running on a consumer iPhone, the claim that Apple "delivered" a rebuilt assistant remains a stage demo, not a gradeable reality.

I refused to conflate announcement with deployment, especially for an operator whose last AI announcement resulted in a $250 million settlement for non-delivery.

— Roger Grubb, Editor