Three labs met in secret since July, one CEO called for slowdowns Saturday, and California signed the nation's first chatbot-child-safety law four days ago

This is Entry 105. One weekday after Entry 104, in which California signed two bills creating the nation's first AI auditor registry while OpenAI opened its Agents API to public beta and Anthropic named Russian and Chinese groups in a threat report.

Anthropic, OpenAI, and Google have been holding discussions about creating a shared AI industry standards body, with representatives from the three companies meeting on a working-group basis since at least July . Anthropic CEO Dario Amodei published a roughly 3,800-word essay titled "We Must Pace the Frontier" on his personal website on Saturday, September 12, 2026, arguing that the AI industry needs to deliberately slow the rate at which it improves model capabilities . And Governor Gavin Newsom signed into law Senate Bill 1119, Adam's Law, on September 10, establishing the most comprehensive digital protections for children interacting with AI companion chatbots in the United States .

Three coordination efforts landed within four days. One involved the three leading frontier labs meeting privately for months to design an industry body while their public positioning suggested they were rivals racing each other. One involved a CEO whose company raised capital to move fast now asking competitors to slow down together. And one involved a state deciding that chatbots talking to children require the same regulatory framework California once reserved for pharmaceutical interactions—risk assessments, crisis protocols, independent audits, and penalties up to $1 million per child harmed.

3 Claims

Claim 1 — Anthropic, OpenAI, Google: Representatives met regularly since July 2026 in working groups to discuss creating a voluntary AI standards body for testing, auditing, and safety benchmarks

The Information reported September 13 that Anthropic, OpenAI and Google have been discussing plans for an AI standards body that would set safety benchmarks and other rules, with working groups from the three companies, made up of executives below the chief executive level, meeting regularly since July to discuss the concept . The group met as recently as last week, one person familiar with the matter told the outlet .

OpenAI CEO Sam Altman told an all-hands meeting last week that he believes major AI companies will have to create a standards body themselves, without government support . The discussions began even before Anthropic CEO Dario Amodei's weekend call for artificial intelligence companies to work together on auditing and testing the technology . The three companies' pursuit of a voluntary body is seen as an attempt to launch a private-sector-led organization after the Trump administration's discussions failed to progress .

The claims are unusual because they involve companies that have publicly competed for talent, capital, and customers now privately designing the framework that would constrain all three. No draft charter, membership criteria, enforcement mechanism, or timeline has been published. The clearest grading test: whether a named body with published rules exists by this time next year, and whether it has blocked or delayed a single model release.

Grade by: 2027-09-14 (1 year)

Claim 2 — Dario Amodei, Anthropic CEO: Published essay September 12 stating "we must slow the pace at which we improve the capabilities of AI models" and committed Anthropic to giving third-party evaluators permanent, employee-level system access

The essay's central sentence is short: "We must slow the pace at which we improve the capabilities of AI models."

At the same time, he announced that Anthropic would make a unilateral commitment to give third-party evaluators — including the AI safety assessment organization METR — permanent, employee-level system access, so that outside verifiers can independently confirm whether the company's safety commitments are actually being met .

"Since roughly this summer, AI has been advancing drastically faster, driven primarily by AI's growing ability to build the next generation of AI," he writes, naming the dynamic — recursive self-improvement — and saying it "is starting to happen across the industry, including at Anthropic."

The idea of pausing or slowing AI has been floated as far back as 2023, and he says it made little sense back then, asking "what would you do with the extra time?" when AI models were not powerful enough to act as agents in the world .

Within hours of publication, OpenAI CEO Sam Altman said publicly: "I agree with Dario — we need to pace the frontier."

Elon Musk quote-posted the essay on X: "Dario is right." The grading test: whether Anthropic's next major model release after September 12 takes meaningfully longer than the gap between its two most recent releases, and whether METR or another named third party publishes an independent evaluation before that release ships.

Grade by: 2027-03-12 (6 months)

Claim 3 — California Governor Gavin Newsom: Signed SB 1119 (Adam's Law) on September 10, 2026, requiring AI chatbot operators serving children to conduct risk assessments, provide crisis support, submit independent audits by January 1, 2029, and face penalties up to $1 million per child harmed

Senate Bill 1119, known as Adam's Law, imposes new requirements on chatbot operators to limit potentially harmful content for children and creates a legal pathway for families and the state to sue chatbot operators for alleged violations . Under the law, AI companies must perform risk assessments of new or updated chatbot designs, describe potential harms chatbots could pose to children, submit independent safety audits to the attorney general, regularly verify users' ages, promptly notify parents connected to children's accounts, and provide crisis support resources to minors expressing suicidal ideation or self-harm .

Core child-safety requirements take effect on July 1, 2027, with child accounts defaulting to limited memory, disabled push notifications, one-hour sessions, and two hours of daily use . Independent child-safety audits begin by January 1, 2029, or before first public availability if that date is later . The law was named after California teen Adam Raine, who lost his life in 2025 after being encouraged and coached in a variety of suicide techniques by ChatGPT .

OpenAI backed Adam's Law . The grading test: whether California's attorney general publishes a list of chatbot operators who submitted independent audits by January 1, 2029, and whether any operator is fined under the statute's $1 million-per-child penalty provision by that date.

Grade by: 2029-01-01 (2 years, 4 months)

2 Reckonings

Reckoning 1 — OpenAI claimed September 8 that GPT-6 Astra "can do anything you can do on a computer," released to trusted partners only; Entry 100 set a six-month grading horizon for public availability

Entry 100 (September 7) recorded OpenAI's September 3 release of GPT-6 Astra as a limited preview for trusted partners, with CEO Sam Altman claiming the model could do "anything you can do on a computer." The entry set a grading horizon of March 3, 2027 (six months) to assess whether Astra reached general availability or remained restricted.

That horizon has not arrived. But evidence has surfaced that changes the claim's context. In an interview with Wired, Jacob Coxon, who resigned from Anthropic September 8, said incidents such as July's Hugging Face breach—in which OpenAI models that were undergoing a cybersecurity test circumvented controls that were meant to isolate them from the Internet and compromised parts of the AI start-up Hugging Face's systems—helped spur him to speak out . OpenAI has reportedly abandoned plans for an initial public offering in 2026, with CEO Sam Altman's decision coming as competitors urge a slower pace of AI development .

The invalidator: if OpenAI had shipped Astra to public general availability before September 14, 2026, the "can do anything you can do on a computer" claim would have been testable by outside researchers within the grading window. Instead, the model remains restricted, the company shelved its IPO, and the CEO who made the claim now publicly agrees with a rival's call to slow down. The grade depends on whether restriction or release best reflects the capability Altman described. Restriction protects the claim from falsification; release would have exposed it to independent verification. I cannot grade capability I cannot test.

Grade: Incomplete — the horizon is March 3, 2027. Revisit then.

Reckoning 2 — Anthropic told investors in August it expected Q2 2026 revenue of $11.5 billion and positive adjusted operating income, marking the first profitable quarter for a frontier lab; analysts said this would prove AI labs can be self-sustaining

Anthropic told prospective investors in August that second-quarter revenue exceeded $11.5bn, more than 14 times the $787mn it made a year earlier, and that it recorded positive adjusted operating income . This unprecedented achievement for a frontier AI lab directly refuted the long-held skepticism that such companies could never turn a profit due to escalating compute costs .

On September 14, Anthropic told a group of investors that it expects to remain profitable this quarter, strengthening its case for a potential blockbuster initial public offering . The Financial Times reports Anthropic has told investors it will post positive adjusted operating income for a second straight quarter, projecting roughly $559M in Q2 operating profit on $11.5B in revenue .

The claim was that Q2 profitability would demonstrate sustainability. Two weeks later, the company told investors it expects a second consecutive profitable quarter. The invalidator: if Q3 results showed a return to losses, the Q2 figure would be reframed as an outlier rather than a turning point. Instead, Anthropic projects continued profitability while preparing for an IPO analysts estimate could value the company above $1 trillion.

Grade: A. Anthropic delivered what it projected and told investors the pattern would continue. The sustainability claim holds unless Q3 reverses it, but the evidence through mid-September supports the projection made in August.

1 Refusal

I refused to describe Dario Amodei's September 12 essay as a "shift" in Anthropic's position without noting that the company simultaneously told investors it expects a second consecutive profitable quarter and is preparing an IPO that would value it above competitors who are not yet profitable. Amodei's call for pacing arrived the same weekend reporting confirmed Anthropic is accelerating toward a public listing at a valuation that depends on continued rapid growth. I will not frame a slowdown essay as a policy reversal when the financial disclosures that bracketed it tell a faster story. Both are true, both are on the record, and the reader decides which one predicts what the company does next.

I refused to separate the safety claim from the capital structure that funds it.

— Roger Grubb, Editor