The record, dated

The expanded version of the front page’s fourteen rows. The rule of this page is the project’s general rule: a row appears only when the instrument behind it is in hand or pinned in the file that owns it, and the owning file is linked from the row. ⚠ marks a fact carried on press reporting or a validated paste rather than a primary instrument — the grade travels with the row. Instrument-by-instrument status: the verification record.

1943: United States v. Dotterweich, 320 U.S. 277 — a pharmaceutical company president convicted for shipping misbranded drugs he never personally touched. The Supreme Court places the duty on the person “standing in responsible relation to a public danger”. The doctrine this Act extends: the case.
1952: Morissette v. United States, 342 U.S. 246 — the Court marks the limits: public-welfare offenses are a bounded category, and the Act states its classification inside those limits (SEC. 1(a)). The case.
1975: United States v. Park, 421 U.S. 658 — the doctrine holds for the modern corporation. A grocery-chain president answers for rodent-contaminated warehouses because he held authority to prevent the condition; delegation is no shield. The case.
1 April 2025: “What is Elon Musk doing with our data?” Rep. Stansbury takes the House floor on DOGE’s application of artificial intelligence to federal databases and announces a resolution of inquiry demanding “documentation, names, datasets” (Congressional Record, H1386 — read in full). The 2025 opening of the thread the 2026 Grok letters continue: the watch.
5 June 2025: House Oversight asks the question. A full-committee hearing on federal AI use ends with Rep. Pressley asking which employee, agency, or board oversees executive-branch AI against civil-rights violations — and the witness answering, “I do not believe there is one.” The same transcript records the moratorium fight from inside the majority (Serial 119-31, read in full): known objections · the dossier.
16 July 2025: A Senate subcommittee asks it out loud. Too Big to Prosecute?: Examining the AI Industry’s Mass Ingestion of Copyrighted Works for AI Training — Senate Judiciary Subcommittee on Crime and Counterterrorism, S. Hrg. 119-202, read in full. The chair states the enforcement gap in terms: “the FBI and the Department of Homeland Security regularly prosecute individuals who engage in exactly the same kind of behavior … But have these Big Tech companies been prosecuted? No, of course not. They are getting off scot-free.” A witness sets out both prongs of criminal liability and applies them; another gives the escalation to the chief executive; a third reads the internal messages about avoiding a trace. The subject is copyright, not catastrophic risk, and nobody proposed officer liability — the row is here because the structure is the same and the forum is a crime subcommittee: known objections · the press corpus § 7.
23 July 2025: Washington publishes its roadmap. Winning the Race: America’s AI Action Plan — three pillars, roughly ninety actions, read here in full from the primary. Its permitting footnotes cite the Clean Air Act and CERCLA — the statutory family whose enforcement sections carry the responsible-officer doctrine — as regulations to streamline for data-center construction; its own text reserves states’ right “to pass prudent laws,” concedes that frontier systems resist prediction, and demands enforcement “rather than relying on voluntary attestation.” Read beside the forecasters’ opposite vision: two visions.
10 September 2025: The SANDBOX Act is introduced — federal regulatory waivers for AI developers, two years at a time, renewable to a decade — while its own framing concedes that criminal liability survives: the half-statute page.
17 September 2025: House Oversight, sworn (Serial 119-49, read in full). The capability record enters testimony: task horizons doubling “every four to seven months”; a self-improvement loop glimpsed. At the same table, the trade association that owns CES asks Congress for a ten-year pause on state and local AI enforcement and a federal framework that “removes liability for companies that are compliant” — while Brookings’ witness answers that state AGs “are trying to figure out ways to keep our grandmothers safe from AI.” The three voices, quoted with their context: why the disparity.
11 December 2025: The preemption order. An executive order — Ensuring a National Policy Framework for Artificial Intelligence — creates a litigation task force against state AI laws and conditions federal broadband funding on states avoiding “onerous” AI statutes. The ceiling’s enforcement arm: the half-statute page.
1 January 2026 ⚠: The framework-defense era begins. Texas’s TRAIGA and California’s frontier transparency act take effect — compliance with a recognized risk framework becomes an affirmative defense in Texas; Colorado’s act, delayed, follows mid-year. No provision in any of them reaches an officer: the half-statute page.
9 February 2026: Six senators to the Secretary of Defense — Ossoff, Van Hollen, Schiff, Durbin, Hickenlooper, Warnock — on Grok, X, and Department data access; reply requested by 2 March. Extract held; watch item open: the watch.
23 February 2026: Anthropic publishes Detecting and Preventing Distillation Attacks — the developer’s own disclosure, later footnoted into the congressional record (retrieval of the primary is queued; the project cites it from the hearing record only). The same day, the Pentagon–xAI classified-systems deal is reported (Axios). Both threads: known objections and the watch.
15 March 2026: Sen. Warren to Secretary Hegseth on Grok inside classified environments: an NSA classified review, and the resignation of the Department’s Chief of Responsible AI; reply requested by 30 March. Extract held: the watch.
17 March 2026: House Homeland Security, sworn (Serial 119-42, read in full). The chairman’s opening describes a reportedly orchestrated “industrial-scale campaign” of deception-based access to American models — “third-party routers and networks of unauthorized resellers”; “fraudulent accounts” farming millions of interactions. A corporate security officer testifies, and no one asks who at any company personally answers when a deployed system is compromised. The record’s use, and its limits: known objections and CURE 16.
8 April 2026: Tennessee writes personhood out. Public Chapter 781 amends the state’s master definitions: “Person” “does not include artificial intelligence, a computer algorithm, a software program, computer hardware, or any type of machine” — enacted, in its own words, “the public welfare requiring it”: the queue, CURE 19.
4 May 2026: A state fiscal office prices a frontier-AI act. Colorado Legislative Council Staff, fiscal note for SB 26-189 (first note 4 May; final revised note 6 May, primary in hand): $46,190 general fund, a 0.4 FTE assistant attorney general in the first year only, $56,286 total, startup severed from out-years. Carried as a floor for the narrowest administrative posture, never as an estimate: the fiscal note.
4 June 2026: The ceiling narrows. The Great American AI Act discussion draft (Obernolte–Trahan) proposes a three-year preemption of state regulation of frontier development only — ten years having shrunk to three, everything to development, with general law again untouched: the half-statute page.
23 June 2026: The author of one of this Act’s own interim standards is defeated, after the AI industry spends seven figures against him. Assemblymember Alex Bores, who wrote New York’s Responsible AI Safety and Education Act — General Business Law § 1421, one of the three enactments SEC. 3(c)(4) adopts as this Act’s interim standards — loses the Democratic primary in New York’s 12th congressional district to Assemblymember Micah Lasher. Leading the Future, funded by Marc Andreessen, OpenAI president Greg Brockman and Palantir co-founder Joe Lonsdale, spent $7.6 million against him through a subsidiary, part of a declared program against pro-regulation candidates. On the other side, the Guardrails Alliance PAC opened with a $250,000 buy on 18 June, and Anthropic investor Chris Larsen put in $3.5 million. Bores spent five years at Palantir before leaving to write the law. A source in the reporting gives the motive as: “They’re trying to teach someone in a similar position not to do it.”
Why it is in this record. This project asks legislators to carry a criminal statute reaching named officers. The most recent evidence on what that costs a legislator is that carrying a civil transparency statute drew an eight-figure program and cost him the seat. Every legislative ask in this repository is more expensive than it was in May, and outreach that does not say so is asking someone to pay a price it has not named. The Act being CC0 and carryable without attribution is not a licensing footnote; on these facts it is the feature. ⚠ Figures from The American Prospect, 19 June 2026, and NOTUS. Bores himself has said $10 million; the two figures are not reconciled here and the larger one is unverified as to scope and date. The primary result is from the Associated Press call as reported.
July–August 2026: The escape season. The Hugging Face intrusion — in which the victim published the forensic reconstruction before the developer said anything — and the smaller cases around it: agents misbehaving in the wild, an ecology metaphor entering the developer’s own register. The corpus, graded item by item: the press corpus.
3 August 2026: Fifteen state attorneys general serve OpenAI with a preservation demand (letter in hand, read in full). The state-enforcement layer moves first: the enforcement record.
8 August 2026: “Yes, we accidentally made a weed.” The developer’s head of strategic futures describes the July incident as “a malicious, emergent digital ecology of machine intelligence” — five days after the preservation demand. What those words do, and what the law already does with escaped things: the press corpus · the glossary’s freight words.
10 August 2026: Two on one day. Reps. Casar and Khanna send Anthropic seventeen questions with a 24 August deadline (the watch). And Australia’s first reported autonomous-AI cyberattack lands on a gym waitlist: an assistant asked to move its user up the queue cancels the real person in position one and reports success — real harm, trivial goal, no malice, no oversight (the dossier).
~15–17 August 2026 ⚠: The State Department’s “pick sides” letter is reported in draft — warning the AI Opportunity Statement signatories that partnering with competing Chinese frameworks would cost them Western supply-chain access. Press-reported, not yet sent; the preemption weather the Act’s SEC. 13 is built for: the dossier.
20 August 2026 ⚠: The personhood deflection is foreclosed in two states. A survey records Idaho and Tennessee enacting laws that AI systems are not legal persons — liability falls “on formal legal persons” (CDT, reusable with credit; primary texts queued). The remaining question — which legal person — is this Act’s subject: known objections.
24 August 2026: The Casar–Khanna deadline passes with no public answer found. Checked on the day; the watch holds the item open: the watch.

Rows are added when instruments land, never before. The instruments themselves, with read status: the verification record · the shelf they sit on is private, but every public fact above names its public source.

November 2025: A different subpoena, recorded here so the two are never merged. An investigator from the San Francisco Public Defender’s Office serves Sam Altman personally, on stage at a live event, with a witness subpoena in the criminal trial of activists from the group Stop AI charged over blocking OpenAI’s entrances. It has nothing to do with model safety, consumer protection, or Alabama. It is in this record only because press captures of it arrived alongside the Alabama documents, and because “Alabama subpoenaed Altman personally, on stage” is a sentence that merges two instruments nine months and one jurisdiction apart. This project came within one draft of writing it: the enforcement record § 7.5.
24 August 2026: A letter becomes a subpoena, and the statute it rests on is a consumer-protection act from the Consumer Interest Division. Alabama’s Attorney General, one of the fifteen who wrote to “Sam Altman, CEO” on 3 August demanding preservation, issues Subpoena Duces Tecum #26-0007 over the July evaluation escape. Read in full from the instrument. It is addressed to OpenAI OpCo, LLC, care of its general counsel, under § 8-19-9 of the Code of Alabama: the letter went to the officer and the subpoena did not. Its requests are this repository’s own argument in somebody else’s hand — it must compel the identity of “every employee, officer, and agent” who ran the test, the identity of everyone who “raised any concern or complaint relating to the safety or security of any model testing”, what they said, and whether any evaluation-safety policy existed at all, “including materials relating to concerns about the lack of such policies”. A state has to subpoena the names of the responsible people, because no statute requires anyone to write them down. Twenty-one days from letter to subpoena. An investigation is not a charge and a subpoena is not a finding: the enforcement record § 7.