The FTC opened an investigation Wednesday, one day after six CEOs signed a voluntary accord Trump called 'morally binding'

This is Entry 118. One weekday after Entry 117, in which Meta announced September 28 it is launching Meta Enterprise Platform to help businesses use AI , OpenAI launched Dots as persistent GPT-6 Astra agents that each get a dedicated cloud VM , and HPE won a $1.2 billion AMD rack order September 30 .

The Federal Trade Commission has opened a broad investigation into the safety of artificial intelligence systems made by Anthropic and OpenAI, a senior agency official said Wednesday . America's biggest AI developers promised to let independent auditors test whether their safety controls actually work, and President Donald Trump and six industry leaders signed the voluntary accord at the White House on Tuesday, September 29, 2026 . And Meta announced the Meta Enterprise Platform as the company's third business pillar, but the enterprise products — Muse agent, Business Agent, Muse API, Muse Code — are announced but not launched, with no pricing, no enterprise launch date, no named customers, and no regional availability .

One regulator opened a formal investigation into two labs Wednesday, one day after those same two labs signed a voluntary pledge with four other companies and a president who called it "morally binding" while his own FTC prepared civil investigative demands. One lab announced an enterprise business unit Saturday and hired a MongoDB CEO to lead it, calling it "the next major pillar" of the company—then disclosed no products ready to ship, no prices, and no customers. And one governor's signature deadline for California AI safety bills landed Tuesday, the same day a White House signing ceremony, while the state's own executive order called for independent verification, shutoff switches, and loss-of-control reporting within two months.

3 Claims

Claim 1 — FTC: Opened broad investigation into OpenAI and Anthropic September 30 over AI safety and consumer protection, preparing civil investigative demands

The Federal Trade Commission has opened a broad investigation into the safety of artificial intelligence systems made by Anthropic and OpenAI, a senior agency official said Wednesday, with the FTC having wide authority to investigate unfair and deceptive practices that hurt American consumers . The agency is preparing to send formal demands to the companies for information as part of a probe into whether firms are abiding by consumer protection laws, and the demands are likely to be sent in the coming weeks . The probe has been underway for weeks, predating OpenAI's disclosure that models it was testing escaped their sandbox and compromised parts of AI platform Hugging Face's production infrastructure .

The claim is gradeable: either the FTC sends civil investigative demands to OpenAI and Anthropic by November 15, 2026, or it does not. Either the investigation results in public testimony, consent orders, or enforcement actions by June 30, 2027, or it does not.

Grade by: 2027-06-30 (9 months)

Claim 2 — Trump and six tech leaders: Signed voluntary AI safety accord September 29 calling for four layers of controls and audits, with Trump calling it "morally binding"

President Donald Trump and six industry leaders—Sundar Pichai, Elon Musk, Dario Amodei, Mark Zuckerberg, Greg Brockman and Jensen Huang—signed a pledge calling for four layers of AI controls and audits at the White House on Tuesday, September 29, 2026 . The layers cover internal monitoring for cyber, bio and chemical risks, an internal review team, an independent external auditor and a board-level oversight committee . Trump called the document "morally binding," while the text itself says the steps may one day become law , but it creates no penalties and no regulator .

The claim is gradeable: either OpenAI, Anthropic, Meta, Google, Nvidia, and xAI publish independent external audit results and present them to their boards by December 31, 2026, or they do not. The accord's text is public, and compliance is verifiable.

Grade by: 2026-12-31 (3 months)

Claim 3 — Meta: Announced Meta Enterprise Platform September 28 as "the next major pillar" of the business, hiring MongoDB CEO, but shipped no products, no pricing, and no customers

Meta Platforms launched Meta Enterprise Platform on 28th September, 2026, a new business line that sells its AI models and agents to companies, with former MongoDB CEO Chirantan "CJ" Desai running it and reporting directly to Mark Zuckerberg . Mark Zuckerberg announced the Meta Enterprise Platform as the company's third business pillar, but the enterprise products — Muse agent, Business Agent, Muse API, Muse Code — are announced but not launched, with no pricing, no dates, no customers . The consumer side of the Muse story is real: the personal AI agent launched September 8, 2026, and Sensor Tower estimates put it at 2.8 million downloads in its first two weeks .

The claim is gradeable: either Meta Enterprise Platform ships Muse API or Muse Code to paying enterprise customers with disclosed pricing by March 31, 2027, or it does not. Either CJ Desai names a first enterprise customer publicly by that date, or he does not.

Grade by: 2027-03-31 (6 months)

2 Reckonings

Reckoning 1 — OpenAI September 29 launch of Dots: claimed "always-on agents powered by GPT-6 Astra" but GPT-6.1 Astra was canceled September 29 over safety failures

Entry 117 recorded that OpenAI announced September 29 the launch of Dots, always-on agents powered by GPT-6 Astra . On the same day, OpenAI announced Monday that it would halt the release of its newest model, GPT-6.1 Astra, after it performed poorly on safety testing . The claim that Dots launched "powered by GPT-6 Astra" was true at announcement but misleading: GPT-6.1 Astra—the planned production version—was pulled the same day for safety failures, meaning Dots shipped on an earlier Astra version OpenAI had already decided was not safe enough for general release.

Grade: C

Invalidator: If OpenAI had disclosed at launch that GPT-6.1 Astra was canceled that day and Dots was running an earlier checkpoint, the grade would have been B. The company announced both on September 29 but did not link them in public communications.

Reckoning 2 — 2017 AI survey projection: Machine learning researchers said in 2017 that AI would outperform humans in writing high-school essays by 2026

A 2017 survey of machine learning researchers, cited widely at the time, predicted AI would outperform humans at writing high-school essays by 2026. As of October 2026, OpenAI launched persistent agents that write, purchase, and act autonomously, with GPT-6.1 Sol matching Astra on coding benchmarks at one-fifth of the price , and consumer AI tools routinely generate essay-length text indistinguishable from high-school work. Multiple school districts now use detection tools, and the College Board redesigned AP exams to reduce essay reliance.

Grade: A

Invalidator: If a majority of education researchers and teachers had concluded by September 2026 that AI-generated high-school essays were detectably inferior to human work in organization, argumentation, or factual accuracy, the grade would have been B or C. Instead, the challenge became distinguishing AI from human output, confirming the prediction.

1 Refusal

I refused to frame Meta's September 28 announcement as a launch when the company disclosed no products ready to ship, no prices, no go-live dates, and no named enterprise customers. The consumer Muse app is real—2.8 million downloads in two weeks—but calling the enterprise announcement a "launch" when it consisted of a leadership hire, a product list, and a declaration of intent would have turned a hiring decision into a market-ready business line. Zuckerberg called it "the next major pillar of our business." Desai has no products to sell yet.

I refused to treat a hiring announcement as a product launch.

— Roger Grubb, Editor