06 Your rules
Plain versionBack to the room
The exhibit is still arriving. It will wake when all of it is here.
1 of 7 The rope
“The velvet rope is the terms of service.”
2 of 7 The label
06
Your rules
Safety Theater How AI Companies Lobbied for the Regulations That Kill Their Competitors
The guardrails have a gatekeeper.
The record begins:
Decided On September 29, 2024, Governor Gavin Newsom returned Senate Bill 1047, which would have required developers of large artificial intelligence models to put safeguards in place, to the California Senate “without my signature.”
Decided The veto: “a false sense of security”
Decided SB 53 signed, September 29, 2025
The fine print
6.01 Decided Veto message on Senate Bill 1047, Office of the Governor of California, dated September 29, 2024 Office of the Governor of California · September 29, 2024 · as of October 16, 2026
The veto message says that by focusing only on the most expensive and largest models, the bill “could give the public a false sense of security.” It also says: “we cannot afford to wait for a major catastrophe to occur before taking action to protect the public.”
6.02 Decided Veto message on Senate Bill 1047, Office of the Governor of California, dated September 29, 2024 Office of the Governor of California · September 29, 2024 · as of October 16, 2026
On September 29, 2025, Governor Gavin Newsom approved Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act. It was filed with the Secretary of State the same day, as Chapter 138.
6.03 Decided SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
Three questions. The record answers in its own words.
Asked?
Reported built on an experts’ report
Reported SB 1047 letter: “signed by over 125 employees”
Pending whose input the law says to take into account
The fine print
The governor’s office says SB 53 “builds on recommendations from California’s first-in-the-nation report,” released by “a group of world-leading AI academics and experts” convened at the governor’s request.
6.04 Reported Governor Newsom signs SB 53, advancing California’s world-leading artificial intelligence industry (press release) Office of the Governor of California · September 29, 2025 · as of October 16, 2026
On September 9, 2024, current and former employees of AI companies including OpenAI, Google DeepMind, Anthropic, Meta and xAI, writing “in our personal capacities,” asked the governor to sign SB 1047. The letter’s page says it was “signed by over 125 employees of frontier AI labs.”
6.05 Reported Lab Employee Statement on Extreme AI Risks (letter to Governor Newsom on SB 1047) Call to Lead (the signatories’ own site) · September 9, 2024 · as of October 17, 2026
By January 1, 2027, and every year after, SB 53 has California’s Department of Technology recommend whether to update who the law covers, taking into account “input from stakeholders, including academics, industry, the open-source community, and governmental entities.”
6.06 Pending SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Business and Professions Code section 22757.14(a)-(b) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
Paid?
Reported Anthropic’s federal lobbying on all issues, 2025: $3,130,000
Reported Google’s federal lobbying on all issues, 2025: $13,100,000
Reported OpenAI’s federal lobbying on all issues, 2025: $2,990,000
The fine print
For 2025, Anthropic’s own reports put its federal lobbying expenses at $360,000, $920,000, $1,010,000 and $840,000, quarter by quarter: $3,130,000 for the year. The second-quarter figure is from an amended report that replaced $910,000.
$360,000 + $920,000 + $1,010,000 + $840,000 = $3,130,0006.07 Reported Anthropic (registrant and client), Lobbying Disclosure Act quarterly reports (LD-2) for 2025, including the second-quarter amendment, as listed by the LDA.gov API LDA.gov, the Lobbying Disclosure Act database of the U.S. Congress (Secretary of the Senate, Office of Public Records; Clerk of the House, Legislative Resource Center) · January 20, 2026 · as of October 17, 2026
For 2025, Google’s own reports, filed by Google Client Services LLC, put its federal lobbying expenses at $3,010,000, $3,170,000, $3,570,000 and $3,350,000, quarter by quarter: $13,100,000 for the year. The figure covers every issue the reports list, from AI to taxes, trade, immigration and defense.
$3,010,000 + $3,170,000 + $3,570,000 + $3,350,000 = $13,100,0006.08 Reported Google Client Services LLC (registrant and client), Lobbying Disclosure Act quarterly reports (LD-2) for 2025, as listed by the LDA.gov API LDA.gov, the Lobbying Disclosure Act database of the U.S. Congress (Secretary of the Senate, Office of Public Records; Clerk of the House, Legislative Resource Center) · January 20, 2026 · as of October 17, 2026
For 2025, OpenAI’s own reports put its federal lobbying expenses at $560,000, $620,000, $920,000 and $890,000, quarter by quarter: $2,990,000 for the year.
$560,000 + $620,000 + $920,000 + $890,000 = $2,990,0006.09 Reported OpenAI OpCo, LLC (registrant and client), Lobbying Disclosure Act quarterly reports (LD-2) for 2025, as listed by the LDA.gov API LDA.gov, the Lobbying Disclosure Act database of the U.S. Congress (Secretary of the Senate, Office of Public Records; Clerk of the House, Legislative Resource Center) · January 20, 2026 · as of October 17, 2026
Anthropic’s own reports put its 2024 federal lobbying expenses at $100,000, $150,000, $220,000 and $250,000, quarter by quarter: $720,000 for the year.
$100,000 + $150,000 + $220,000 + $250,000 = $720,0006.10 Reported Anthropic (registrant and client), Lobbying Disclosure Act registration (LD-1) and quarterly reports (LD-2) for 2024, as listed by the LDA.gov API LDA.gov, the Lobbying Disclosure Act database of the U.S. Congress (Secretary of the Senate, Office of Public Records; Clerk of the House, Legislative Resource Center) · January 21, 2025 · as of October 17, 2026
Google’s own quarterly lobbying reports, filed by Google Client Services LLC, put its federal lobbying expenses for 2024 at $3,080,000, $2,960,000, $3,200,000 and $3,070,000: $12,310,000 for the year.
$3,080,000 + $2,960,000 + $3,200,000 + $3,070,000 = $12,310,0006.11 Reported Google Client Services LLC (registrant and client), Lobbying Disclosure Act quarterly reports (LD-2) for 2024, as listed by the LDA.gov API LDA.gov, the Lobbying Disclosure Act database of the U.S. Congress (Secretary of the Senate, Office of Public Records; Clerk of the House, Legislative Resource Center) · January 21, 2025 · as of October 17, 2026
OpenAI’s own reports put its federal lobbying expenses for 2024 at $340,000, $480,000, $430,000 and $510,000, quarter by quarter: $1,760,000 for the year. The second-quarter figure is from an amended report that replaced $460,000.
$340,000 + $480,000 + $430,000 + $510,000 = $1,760,0006.12 Reported OpenAI OpCo, LLC (registrant and client), Lobbying Disclosure Act quarterly reports (LD-2) for 2024, including the second-quarter amendment, as listed by the LDA.gov API LDA.gov, the Lobbying Disclosure Act database of the U.S. Congress (Secretary of the Senate, Office of Public Records; Clerk of the House, Legislative Resource Center) · January 21, 2025 · as of October 17, 2026
TechCrunch reported on January 24, 2025 that, “according to data compiled by OpenSecrets, 648 companies spent on AI lobbying in 2024 versus 458 in 2023”: 190 more companies, about 41 percent more.
648 − 458 = 190 more companies(648 − 458) ÷ 458 × 100 = 41.5 percent more6.13 Reported AI companies upped their federal lobbying spend in 2024 amid regulatory uncertainty (by Kyle Wiggers) TechCrunch · January 24, 2025 · as of October 16, 2026
In the Lobbying Disclosure Act’s terms, a “lobbying contact” is a communication to a covered official made on behalf of a client; “lobbying activities” also take in the planning, research and background work behind those contacts. A “lobbying report” is the quarterly report, form LD-2.
6.14 Reported Lobbying Disclosure Act Guidance (last revised February 28, 2025), Section 3 (definitions), page 3 Secretary of the Senate and Clerk of the U.S. House of Representatives · February 28, 2025 · as of October 16, 2026
The officials these reports cover are federal ones: among them the President, the Vice President, the staff of the Executive Office of the President, officials in Executive Level I to V posts, and Members of Congress and their staff.
6.61 Reported Lobbying Disclosure Act Guidance (last revised February 28, 2025), Section 3 (definitions), page 3 Secretary of the Senate and Clerk of the U.S. House of Representatives · February 28, 2025 · as of October 17, 2026
The expenses a company with its own lobbyists reports are its own costs: “employee compensation, office overhead, or payments to vendors, which may include lobbying firms,” in the guidance’s words. A quarter’s total of $5,000 or more must be reported as “a good faith estimate,” rounded to the nearest $10,000.
6.62 Reported Lobbying Disclosure Act Guidance (last revised February 28, 2025), Section 6 (income or expense recording), page 16 Secretary of the Senate and Clerk of the U.S. House of Representatives · February 28, 2025 · as of October 17, 2026
Under the federal disclosure rules, a company with its own lobbyists must report all its lobbying expenses, “including all payments to retained lobby firms.” Its reported expense “should be greater than the fees reported by the lobbying firm” it hires.
6.15 Reported Lobbying Disclosure Act Guidance (last revised February 28, 2025), Section 6 (income or expense recording), page 16 Secretary of the Senate and Clerk of the U.S. House of Representatives · February 28, 2025 · as of October 16, 2026
Told?
Decided a report with each new frontier model
Decided publish a safety framework
The fine print
Before or when it deploys a new frontier model, a large frontier developer must publish a transparency report summarizing its assessments of the model’s catastrophic risks, their results and “the extent to which third-party evaluators were involved.”
6.16 Decided SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Business and Professions Code section 22757.12(c) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
SB 53 requires each large frontier developer to write, follow and publish on its website a “frontier AI framework,” which the law defines as “documented technical and organizational protocols to manage, assess, and mitigate catastrophic risks,” and to review it at least once a year.
6.17 Decided SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Business and Professions Code sections 22757.11(g) and 22757.12(a)-(b) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
In SB 53, a “frontier model” is a foundation model trained with more than 10^26 operations of computing power, and a “large frontier developer” is one that, with its affiliates, had more than $500,000,000 in gross revenue the year before.
6.18 Decided SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Business and Professions Code section 22757.11(i)-(j) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
In SB 53, “catastrophic risk” is a foreseeable and material risk that a developer’s frontier model (in the whistleblower rules, a foundation model) will materially contribute to death or serious injury to over 50 people, or over $1 billion in property damage or loss, in one incident of a listed kind, with exclusions.
6.19 Decided SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Business and Professions Code section 22757.11(c) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 17, 2026
3 of 7 The number
In July 2025 the U.S. Senate voted on striking a 10-year pause on state AI laws, tied to new broadband money, from a budget bill. How many of the 100 senators voted to strike it? Guess first.
You guessed
Decided 99 to 1
Withdrawn the 10-year “Temporary Pause”
Decided a task force to challenge state AI laws
The fine print
At 4:08 a.m. on July 1, 2025, the U.S. Senate voted 99 to 1 for an amendment “to strike the section relating to support for artificial intelligence” from its substitute text for H.R. 1.
6.20 Decided Roll Call Vote 119th Congress - 1st Session, Vote Number 363 (On the Amendment, S.Amdt. 2814 to S.Amdt. 2360 to H.R. 1) United States Senate (Senate Legislative Information System) · July 1, 2025 · as of October 16, 2026
The struck section 40012 would have added $500 million to the federal broadband program (BEAD) and opened it to AI infrastructure. A state receiving the money could not enforce, “during the 10-year period” after enactment, its laws “limiting, restricting, or otherwise regulating” AI models and systems, with exceptions.
6.21 Withdrawn S.Amdt. 2360 to H.R. 1, 119th Congress (amendment in the nature of a substitute, as submitted June 28, 2025), section 40012 Congress.gov (Library of Congress), text from the Congressional Record · June 28, 2025 · as of October 16, 2026
An executive order signed on December 11, 2025 told the Attorney General to set up an AI Litigation Task Force “whose sole responsibility shall be to challenge State AI laws” at odds with its policy of “a minimally burdensome national policy framework for AI.”
6.22 Decided Executive Order 14365 of December 11, 2025, Ensuring a National Policy Framework for Artificial Intelligence (Federal Register, Vol. 90, No. 239, December 16, 2025, pages 58499-58501), full text Office of the Federal Register, National Archives (Federal Register) · December 16, 2025 · as of October 16, 2026
The order asks two White House advisers for “a legislative recommendation establishing a uniform Federal policy framework for AI that preempts State AI laws that conflict” with it, but not state laws on child safety, AI data center infrastructure, or state government procurement and use of AI.
6.23 Proposed Executive Order 14365 of December 11, 2025, Ensuring a National Policy Framework for Artificial Intelligence (Federal Register, Vol. 90, No. 239, December 16, 2025, pages 58499-58501), full text Office of the Federal Register, National Archives (Federal Register) · December 16, 2025 · as of October 16, 2026
In March 2026 the White House recommended that “Congress should preempt state AI laws that impose undue burdens,” but not states’ zoning, their own use of AI, or general laws such as those “to protect children, prevent fraud, and protect consumers,” and that states “should not be permitted to regulate AI development.”
6.52 Proposed National Policy Framework for Artificial Intelligence: Legislative Recommendations (March 2026), page 4 The White House · March 20, 2026 · as of October 17, 2026
Preemption: the Constitution makes federal law “the supreme Law of the Land,” and courts call it preemption when federal law displaces state law. In the Library of Congress’s words, “Federal law expressly preempts state law when it contains explicit language to that effect.”
6.24 Reported Constitution Annotated, ArtVI.C2.3.4 Modern Doctrine on Supremacy Clause Library of Congress (Constitution Annotated, Congress.gov) · October 16, 2026 · as of October 16, 2026
4 of 7 Their side
The companies’ own words, in their letters, a post and an essay, and as they gave them to reporters.
Reported “benefits likely outweigh its costs”
Reported “won’t wait for consensus in Washington”
Reported “far too blunt an instrument”
The fine print
Anthropic’s chief executive wrote to the governor on August 21, 2024 that the amended SB 1047 was “substantially improved, to the point where we believe its benefits likely outweigh its costs. However, we are not certain of this, and there are still some aspects of the bill which seem concerning or ambiguous to us.”
6.25 Reported Letter from Dario Amodei, Chief Executive Officer, Anthropic, PBC, to Governor Gavin Newsom, "RE: SB 1047 (Wiener)", August 21, 2024 Anthropic, PBC · August 21, 2024 · as of October 16, 2026
Anthropic endorsed SB 53 on September 8, 2025, writing that “frontier AI safety is best addressed at the federal level instead of a patchwork of state regulations,” but that “powerful AI advancements won’t wait for consensus in Washington.”
6.26 Reported Anthropic is endorsing SB 53 (announcement, September 8, 2025) Anthropic, PBC · September 8, 2025 · as of October 16, 2026
In a New York Times essay on June 5, 2025, Anthropic’s chief executive called the motives behind a proposed 10-year moratorium on state AI regulation “understandable,” but the moratorium itself “far too blunt an instrument.”
6.27 Reported Anthropic C.E.O.: Don’t Let A.I. Companies off the Hook (guest essay by Dario Amodei), The New York Times, June 5, 2025 (Internet Archive capture of June 5, 2025) The New York Times (opinion); copy in the Internet Archive · June 5, 2025 · as of October 16, 2026
Anthropic’s August 2024 letter on SB 1047 listed, among issues it said were not addressed, “overly expansive whistleblower protections that are subject to abuse.” The burdens of those provisions, it said, were “likely to be manageable, if the executive branch takes a judicious approach to implementation.”
6.60 Reported Letter from Dario Amodei, Chief Executive Officer, Anthropic, PBC, to Governor Gavin Newsom, "RE: SB 1047 (Wiener)", August 21, 2024 Anthropic, PBC · August 21, 2024 · as of October 17, 2026
Reported “rather than a patchwork of state laws”
Reported OpenAI’s ask: count federal and EU deals
Reported “a critical path toward harmonization”
The fine print
On August 21, 2024, OpenAI wrote to SB 1047’s author opposing it: “A federally-driven set of AI policies, rather than a patchwork of state laws, will foster innovation.” It backed “the intent behind some of its provisions” and saw “a role for states to play” on issues like potential bias in hiring and deepfakes.
6.28 Reported Letter from Jason Kwon, Chief Strategy Officer, OpenAI, to Senator Scott Wiener, "Re: Senate Bill 1047 (Wiener) - Safe and Secure Innovation for Frontier Artificial Intelligence Models Act - OPPOSE", August 21, 2024 OpenAI (copy of the signed letter published by Maginative) · August 21, 2024 · as of October 17, 2026
On August 11, 2025, OpenAI asked California’s governor to treat frontier developers as compliant with state requirements when they sign a parallel framework, such as the EU’s Code of Practice, or a safety agreement with a federal agency. The letter does not name SB 53.
6.29 Reported Letter from Chris Lehane, Chief Global Affairs Officer, OpenAI, to Governor Gavin Newsom, "Re: Recognition of International and Federal AI Safety Frameworks for State Law Compliance", August 11, 2025 OpenAI · August 11, 2025 · as of October 16, 2026
Fortune reported on September 30, 2025 that OpenAI, which had not endorsed SB 53, told news outlets after the signing that it was “pleased to see that California has created a critical path toward harmonization with the federal government—the most effective approach to AI safety.”
6.30 Reported California governor signs landmark AI safety law, forcing major tech companies to disclose protocols and protect whistleblowers (by Beatrice Nolan), Fortune, September 30, 2025 (Internet Archive capture of April 6, 2026) Fortune (copy in the Internet Archive) · September 30, 2025 · as of October 16, 2026
Reported “a positive step in that direction”
The fine print
Fortune reported on September 30, 2025 that a Meta spokesperson told media the company “supports balanced AI regulation” and called SB 53 “a positive step in that direction.”
6.31 Reported California governor signs landmark AI safety law, forcing major tech companies to disclose protocols and protect whistleblowers (by Beatrice Nolan), Fortune, September 30, 2025 (Internet Archive capture of April 6, 2026) Fortune (copy in the Internet Archive) · September 30, 2025 · as of October 16, 2026
5 of 7 What fair looks like
What has been made law, put off, changed or signed up to so far. Each line carries the exact word for where it stands.
Decided up to $1,000,000 a violation
Decided 15 days to report; 24 hours if imminent
Decided protection for employees who report
Reported a portal for incident reports
The fine print
Since January 1, 2026, a large frontier developer that fails to publish or send SB 53’s required documents, report a critical safety incident or follow its own framework can face a civil penalty of up to $1,000,000 per violation, in a suit “brought only by the Attorney General.”
6.32 Decided California Business and Professions Code section 22757.15 (Transparency in Frontier Artificial Intelligence Act), as displayed by California Legislative Information California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
SB 53 gives a frontier developer 15 days from discovery to report a critical safety incident to California’s Office of Emergency Services. If the incident poses an imminent risk of death or serious physical injury, the developer must tell an authority such as a law enforcement or public safety agency within 24 hours.
6.33 Decided SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Business and Professions Code section 22757.13(c) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
SB 53 bars frontier developers from using rules, policies or contracts to stop employees who handle safety risks from reporting to the Attorney General or a federal authority a specific and substantial danger to public health or safety from a “catastrophic risk,” or a violation of the Act, and from retaliating for it.
6.34 Decided SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Labor Code sections 1107(b) and 1107.1(a) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
California’s Office of Emergency Services runs a portal where “members of the public and frontier developers can confidentially submit” SB 53 reports: critical safety incidents, and periodic summaries of assessments of catastrophic risk from internal use of frontier models.
6.35 Reported Transparency in Frontier AI Act Reporting (California Cybersecurity Integration Center page) California Governor’s Office of Emergency Services · October 16, 2026 · as of October 16, 2026
SB 53 lets the Office of Emergency Services name federal rules with incident-reporting standards “substantially equivalent to, or stricter than” its own. A developer that declares it follows one “shall be deemed in compliance” with the state’s reporting section; failing the federal rule then breaks state law.
6.36 Decided SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Business and Professions Code section 22757.13(h)-(i) California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
SB 53 also creates a consortium to design “CalCompute,” a public cloud computing cluster, with a report due to the Legislature by January 1, 2027. That part takes effect “only upon an appropriation in a budget act, or other measure, for the purposes of this section.”
6.37 Pending SB-53 Artificial intelligence models: large developers (2025-2026), chaptered text (Chapter 138, Statutes of 2025), Government Code section 11546.8 California Legislative Information (California State Legislature) · September 29, 2025 · as of October 16, 2026
Reported OpenAI, 2024: most ex-staff released, Bloomberg reported
Reported OpenAI’s exit papers, as Vox reported
The fine print
Bloomberg reported on May 23, 2024 that OpenAI was releasing most former employees from non-disparagement agreements, “unless the nondisparagement provision was mutual,” would remove the language from its exit paperwork, and had told former staff it “has not canceled, and will not cancel, any vested units.”
6.53 Reported OpenAI Releases Former Staffers From Non-Disparagement Clauses (by Shirin Ghaffary and Rachel Metz, Bloomberg News), as carried by BNN Bloomberg (Internet Archive capture of May 24, 2024) Bloomberg News (via BNN Bloomberg; copy in the Internet Archive) · May 23, 2024 · as of October 17, 2026
Vox reported on May 17, 2024 that OpenAI’s exit agreement for departing staff “forbids them, for the rest of their lives, from criticizing their former employer,” and that declining or breaking it could cost all vested equity. OpenAI told Vox it had “never canceled any current or former employee’s vested equity.”
6.54 Reported ChatGPT can talk, but OpenAI employees sure can’t (by Kelsey Piper; updated May 18, 2024) Vox · May 17, 2024 · as of October 17, 2026
Vox’s update of May 18, 2024 says OpenAI’s chief executive acknowledged that the company’s off-boarding documents had contained a provision about “potential equity cancellation” and said the company was changing that language.
6.55 Reported ChatGPT can talk, but OpenAI employees sure can’t (by Kelsey Piper; updated May 18, 2024) Vox · May 18, 2024 · as of October 16, 2026
On May 18, 2024, OpenAI’s chief executive posted: “we have never clawed back anyone’s vested equity, nor will we do that if people do not sign a separation agreement (or don’t agree to a non-disparagement agreement). vested equity is vested equity, full stop.”
6.56 Reported Post by Sam Altman (@sama) on X, May 18, 2024, as returned by X’s public embed service (oEmbed) X (post by Sam Altman, OpenAI) · May 18, 2024 · as of October 16, 2026
Reported Anthropic, 2024: removing default non-disparagement, it said
Reported Anthropic, 2024: signers may now say they signed, it said
The fine print
In July 2024 an Anthropic cofounder wrote that the company had “historically included standard non-disparagement terms by default in severance agreements, and in some non-US employment contracts,” and that since June 1 it had been “going through our standard agreements and removing these terms.”
6.57 Reported Comment by Sam McCandlish, an Anthropic cofounder, on LessWrong, posted July 4, 2024 (04:26 UTC) LessWrong (a comment by Anthropic’s cofounder Sam McCandlish) · July 4, 2024 · as of October 17, 2026
The cofounder added: “We’re not claiming that Anthropic never offered a confidential non-disparagement agreement.” Signers were now “free to talk about having signed,” the comment said; Anthropic would “continue to honor all of Anthropic’s non-disparagement and non-disclosure obligations, e.g. from mutual agreements.”
6.58 Reported Reply by Sam McCandlish, an Anthropic cofounder, on LessWrong, posted July 9, 2024 (01:43 UTC) LessWrong (a comment by Anthropic’s cofounder Sam McCandlish) · July 9, 2024 · as of October 17, 2026
The cofounder also wrote that Anthropic had “never tied non-disparagement agreements to vested equity,” that anyone who signed one “is free to state that fact,” and that it “will not enforce the non-disparagement agreement” against a signer who raises concerns about safety at Anthropic.
6.59 Reported Comment by Sam McCandlish, an Anthropic cofounder, on LessWrong, posted July 4, 2024 (04:26 UTC) LessWrong (a comment by Anthropic’s cofounder Sam McCandlish) · July 4, 2024 · as of October 16, 2026
Decided moved to December 2, 2027
Decided why the dates moved
Decided more power for the AI Office
The fine print
The regulation moves the start of the AI Act’s rules for high-risk systems in the areas listed in Annex III from August 2, 2026 to December 2, 2027, and sets August 2, 2028 for high-risk systems covered under Annex I. The rules stay in the Act; their start dates moved.
6.38 Decided Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI), recital 40 and Article 1(40) EUR-Lex (Publications Office of the European Union) · July 24, 2026 · as of October 16, 2026
The regulation gives its reason for the delay: “the delayed availability of standards, common specifications, and alternative guidance and the delayed establishment of national competent authorities,” which it says risked “a significant increase in implementation costs.”
6.39 Decided Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI), recital 40 EUR-Lex (Publications Office of the European Union) · July 24, 2026 · as of October 16, 2026
It makes the EU’s AI Office “exclusively competent” to supervise and enforce the AI Act for AI systems built on a general-purpose AI model by the same provider or company group, with exceptions, and gives it “all the powers of a market surveillance authority” for them.
6.40 Decided Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI), Article 1, points (31) and (32) (new Article 75(1) and Article 75a of the AI Act) EUR-Lex (Publications Office of the European Union) · July 24, 2026 · as of October 16, 2026
The EU’s amending regulation 2026/1744, the “Digital Omnibus on AI,” signed on July 8, 2026, was published in the Official Journal on July 24, 2026 and entered into force on July 27, 2026.
6.41 Decided Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI), EUR-Lex document information EUR-Lex (Publications Office of the European Union) · July 24, 2026 · as of October 16, 2026
The AI Act’s Annex III lists the uses of AI that count as high-risk, with some exceptions, in areas including biometrics, critical infrastructure, education, employment, access to essential services such as credit, and law enforcement.
6.42 Decided Regulation (EU) 2024/1689 (Artificial Intelligence Act), as published in the Official Journal, Annex III EUR-Lex (Publications Office of the European Union) · July 12, 2024 · as of October 16, 2026
In the EU’s AI Act, a “general-purpose AI model” is one that “displays significant generality and is capable of competently performing a wide range of distinct tasks,” and that can be built into many other systems and applications.
6.43 Decided Regulation (EU) 2024/1689 (Artificial Intelligence Act), as published in the Official Journal, Article 3(63) EUR-Lex (Publications Office of the European Union) · July 12, 2024 · as of October 16, 2026
Reported “a voluntary tool”
Decided “limited legal effect”
Reported 22 signatories
The fine print
The European Commission published the General-Purpose AI Code of Practice on July 10, 2025. It calls the code “a voluntary tool, prepared by independent experts in a multi-stakeholder process,” and says companies that sign it “can show they comply with the AI Act by adhering to the code.”
6.44 Reported The General-Purpose AI Code of Practice (Shaping Europe’s digital future; page last updated 7 October 2026) European Commission (Directorate-General for Communications Networks, Content and Technology) · October 7, 2026 · as of October 16, 2026
The EU’s 2026 amending regulation says such codes “have limited legal effect, and in particular do not grant a presumption of conformity,” while providers “should be able to rely” on codes of practice that the Commission has assessed as adequate.
6.45 Decided Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI), recital 41 EUR-Lex (Publications Office of the European Union) · July 24, 2026 · as of October 16, 2026
On October 16, 2026, the Commission’s page listed 22 signatories of the code, among them Amazon, Anthropic, Google, IBM, Microsoft, Mistral AI and OpenAI. xAI signed only the code’s Safety and Security chapter.
6.46 Reported The General-Purpose AI Code of Practice: "Signatories of the code of practice" (Shaping Europe’s digital future; page last updated 7 October 2026) European Commission (Directorate-General for Communications Networks, Content and Technology) · October 7, 2026 · as of October 16, 2026
Under the AI Act, general-purpose AI providers “may rely on codes of practice” to show they meet their duties until a harmonised standard is published. It is compliance with European harmonised standards that “grants providers the presumption of conformity.”
6.47 Decided Regulation (EU) 2024/1689 (Artificial Intelligence Act), as published in the Official Journal, Article 53(4) EUR-Lex (Publications Office of the European Union) · July 12, 2024 · as of October 16, 2026
Reported “up to $10 million” for a safety institute
Reported renamed in 2025
Reported renamed again in 2026
The fine print
On March 7, 2024, the Senate majority leader announced that the fiscal 2024 Commerce, Justice and Science spending bill included “up to $10 million for establishment of the U.S. Artificial Intelligence Safety Institute” at NIST, the National Institute of Standards and Technology.
6.48 Reported Majority Leader Schumer Announces First-Of-Its-Kind Funding To Establish A U.S. Artificial Intelligence Safety Institute (press release) Office of Senate Majority Leader Chuck Schumer (press release) · March 7, 2024 · as of October 16, 2026
On June 3, 2025, the Commerce Department announced plans to reform “the agency formerly known as the U.S. AI Safety Institute” into the Center for AI Standards and Innovation, to serve as “industry’s primary point of contact within the U.S. Government” for testing and research.
6.49 Reported Statement from U.S. Secretary of Commerce Howard Lutnick on Transforming the U.S. AI Safety Institute into the Pro-Innovation, Pro-Science U.S. Center for AI Standards and Innovation U.S. Department of Commerce, Office of Public Affairs · June 3, 2025 · as of October 16, 2026
On October 16, 2026, NIST’s page for the center named it the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), which “will serve as industry’s primary point of contact within the U.S. government to facilitate testing and collaborative research.”
6.50 Reported Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), NIST web page National Institute of Standards and Technology (NIST), U.S. Department of Commerce · October 16, 2026 · as of October 16, 2026
Decided yearly independent audits from 2028, by state law
The fine print
Illinois’ Artificial Intelligence Safety Measures Act takes effect January 1, 2027. From January 1, 2028, each large frontier developer must hire a third party every year for an independent audit of its compliance with the Act’s framework rules, and may not tie the auditor’s pay to the results.
6.51 Decided Illinois Public Act 104-0538 (SB 315, 104th General Assembly), the Artificial Intelligence Safety Measures Act, enrolled text, Sections 1, 10(d) and 99 Illinois General Assembly · July 6, 2026 · as of October 16, 2026
6 of 7 The stamp
Yours to give. There is no right answer, and the museum keeps no count.
In the room, this is where you stamp your ticket with one of five stamps. The plain version asks nothing of you.
Kept on this device only. Nothing is sent.
7 of 7 Three doors
Three ways on. All three lead to the series’ own site.
From the book · Safety TheaterThe model has no opinions. The lobbying budget has several.