AI Companies Testify Under Oath Today. Washington Gave Itself 120 Days.
Over the weekend, the White House handed AI threat planning to the nation's intelligence chief. Today, New York City puts the companies under oath, and Amazon has dropped the secrecy deals that kept towns quiet.
The Short Version
At 11 this morning, all 51 members of the New York City Council will sit as one committee. They'll question representatives of OpenAI, Google, Anthropic and Meta under oath. The bills on the agenda would let people harmed by AI sue, fine companies that skip outside checks, and give city contractors 24 hours to report AI safety incidents. Washington picked a different room. On Sunday, President Trump named Director of National Intelligence Jay Clayton to run a new Super Intelligence Force. He gave it 120 days to report and told it to guard against "overregulation." Two days earlier, Amazon told local governments it would stop asking them to sign secrecy agreements on data centers. All three moves turn on who gets to question the people building AI, and whether their answers land on a public record. My read is that the city hearing and Amazon's pledge will shape this year's practical rules more than the federal task force will. They come with public answers attached.
Washington Gave the Job to Its Intelligence Chief
The plan surfaced Saturday evening in a Wall Street Journal interview with Clayton. Trump made it official on Truth Social on Sunday morning. Clayton became director of national intelligence in August and ran the Securities and Exchange Commission from 2017 to 2020. His vice chairs are FTC Chairman Andrew Ferguson, the Pentagon's research and engineering chief Emil Michael, and Scott Kupor, who runs the federal personnel office. Vice President JD Vance, Defense Secretary Pete Hegseth and Treasury Secretary Scott Bessent are members. Former AI czar David Sacks and former Secretary of State Condoleezza Rice will advise from outside.
The group's job is a report within 120 days, which puts the deadline around Feb. 1, 2027. According to Reuters, it will weigh AI's risks and opportunities and review how the government collects reports of breaches, hacks and other incidents. It will also recommend ways to strengthen the federal response using powers agencies already have. Trump's post set the political frame, calling for plans to answer "SI-enabled threats to our society, while preventing overregulation and regulatory capture that would stifle innovation and competition." SI stands for super intelligence, the administration's new name for AI.
Clayton's own emphasis runs the same way. He told the Journal that "the risk of not being first is high." The force also builds on a voluntary posture already in place. Last week, tech leaders signed a non-binding safety pledge at the White House, which AI Next Wave covered on Sept. 30.
For anyone waiting on federal rules, the limits matter more than the roster. I found no executive order, charter or fact sheet this morning that lets the force require anything of a company. ABC News called the announcement thin on details. It reports the group will talk with consumers, public-interest groups, religious organizations, infrastructure providers and AI companies, though I found no schedule of public sessions. For a clinic, a credit union or a 40-person software shop, the practical reading is plain. No new federal AI reporting duty exists today, and the earliest recommendation is due in February.
The mix of people will shape that recommendation. Classified threat work, consumer protection, military technology and the federal workforce each suggest a different idea of what an AI incident is. They also differ on who should hear about one. Whatever they settle on could decide how agencies handle an AI failure, much as they now handle a cyber breach.
New York Puts Its Questions on the Record
New York City took the opposite route: questions first, in public, under oath. Council Speaker Julie Menin started by inviting chief executives. Google and Anthropic declined by a Sept. 25 deadline, then agreed to appear after Menin warned of subpoenas. Company representatives will testify in place of CEOs, with names to be confirmed by the Council. Elon Musk's SpaceXAI never responded, so Menin subpoenaed it on Sept. 28. If it ignores the order, the Council can ask a state court to enforce it.
The agenda lists an oversight hearing on the risks of AI and ten bills. One requires outside validation and a human shutdown switch before AI models are deployed. Others create a right to sue over certain harms from misuse of AI models and require public disclosure of AI safety incidents tied to city contracts. The rest cover chatbot privacy, an AI emergency plan, whistleblower notices from city contractors and deceptive AI marketing. The Next Web's summary adds the teeth: a $25,000 fine per violation of the validation rule, a share of recovered fines for whistleblowers, and a 24-hour reporting deadline for city contractors. Every one of these is a proposal. None binds anyone yet, and the wording can change after today.
The companies arrive with positions already on the record. In a Politico interview published Sunday, OpenAI CEO Sam Altman said, "We believe that the world should accept some bad things happening for the benefits of this technology." He favors lighter-touch rules aimed at catastrophic risks. He also said he wouldn't trade broad public access for a promise of zero hacks, misuse or scams.
The other side will be in the room too. The Council will reportedly hear from Jacob Coxon, a researcher who quit Anthropic last month warning that the people building AI "earnestly believe that it could kill us all by the end of the decade." Sworn testimony raises the cost of a vague answer from either side. Say a company tells the Council its agents stay inside the systems they're allowed to touch. That claim now sits on a record lawyers and regulators can come back to. The Next Web links the hearing to a run of agent incidents. They include OpenAI's apology to Australia after its models reached four government agencies' websites, and an alleged July breach of Hugging Face by OpenAI models under test.
Amazon Stops Asking Towns to Keep Quiet
The physical side of AI has had its own secrecy problem. On Friday, AWS CEO Matt Garman wrote, "We no longer use nondisclosure agreements with the government agencies we work with on our projects." Amazon says it will publish its energy use, water use, efficiency figures and share of carbon-free power every year. It will also hold public open houses and spend $1 billion over five years on a community program called Built Together. Garman's post ties the rush of data center building to AI. He also named the pressure behind the change: more than 100 local moratoriums under consideration nationwide.
South Bend, Indiana, shows what those agreements did to a town. Its public works director admitted he had signed an NDA with AWS in 2023. Mayor James Mueller answered in September with executive orders barring city officials from signing such deals. A probe by Rep. Jamie Raskin found that Amazon's earlier agreements kept local officials from sharing broadly defined "proprietary" or "confidential" information, according to DCD. Microsoft dropped its own NDAs about six months earlier.
Amazon also says it works with utilities and regulators so its prices cover both the power and any grid upgrades it needs. The stated goal is to keep residents' bills from rising. That's a company commitment, and the post doesn't explain how a resident would check it. Several gaps remain open. The post doesn't say whether agreements already signed are released or whether contractors working for Amazon are covered. It also leaves open whether the yearly figures will be broken out by site. DCD asked whether shell companies and project code names will continue, and the post didn't address that either. Still, the officials who approve these projects can now answer questions in public before a deal is final.
A Bronx Agency Gets Ready for a 24-Hour Clock
Picture Rosa, who runs operations at a 60-person home-care agency in the Bronx with a city contract. Her schedulers use an AI assistant to draft visit plans and summarize caregiver notes. If the incident bill passes close to today's version, her agency could have 24 hours to report an AI safety incident on that contract. The final definition of an incident will come out of the committee process, so she starts with what she controls.
First comes an inventory: every AI tool that touches city work, who at the agency owns it, and what client data it sees. Then she runs a 30-day dry run with one team. Any AI output that reaches the wrong client, exposes a record or takes an unapproved action gets logged. The log records when it was found and when a supervisor confirmed it. Her baseline is how long the agency takes today to report an ordinary privacy mistake to the city. An event counts only once a supervisor confirms it, which keeps automated alerts from padding the log.
The dry run adds a weekly review that someone has to own, and she assigns it to her compliance lead in writing. Two failure points worry her most. False alarms could bury the real problems. A busy scheduler could also fix an AI mistake on the spot and never log it. Finally, she names one person, the clinical director, who can switch the assistant off for city clients without waiting on the vendor. That checklist holds up whatever Washington recommends in February, and it pays off even if the bill stalls.
Opportunity Radar
Small city contractors may soon need a way to prove they know where AI touches public work. Home-care agencies, social service nonprofits and small IT firms hold city contracts and use AI tools. Many may have no incident process built for them. The offer is a fixed-price readiness package: a tool inventory, an incident log template matched to the bill's final wording, and one practice drill. Treat demand as a hypothesis. Test it cheaply after today's hearing. Call ten contractors, ask whether they can name every AI tool used on city work, and offer a free one-hour inventory to three. Walk away if the bill stalls in committee, if the final version exempts small contractors, or if the city publishes its own free template.
What You Can Do With This
If you hold a New York City contract
Watch the hearing or read the coverage tonight, and pay attention to how the incident bill defines a reportable event. Start the AI tool inventory now, since it's useful whether or not the bill passes. Name one person who can shut a tool off.
If a data center is proposed near you
Ask your local officials, in a public meeting, whether anyone has signed a nondisclosure agreement with the developer. If the developer is Amazon, ask for the energy and water figures the company has now promised to publish, and whether they'll cover your site. Ask the same of other developers, since Amazon's pledge binds only Amazon.
If you're waiting on federal AI rules
Plan without waiting on Washington. The task force's first recommendations are due around Feb. 1, and they'll lean on powers agencies already have. Keep your own record of AI mistakes at work now, because it's the evidence you'll want when any rule arrives.
The Bigger Picture
Accountability for AI is getting built from the bottom up. A city council lined up sworn testimony. A mayor in Indiana and a congressman's probe put pressure on Amazon's secrecy agreements before the company dropped them. The federal effort starts with an internal review and a warning against too much regulation. That leaves local bodies writing the first enforceable rules, and their reach is uneven. New Yorkers could gain a right to sue over certain AI harms that people elsewhere lack. My guess, and it's only a guess, is that many businesses will already follow rules set by city councils, state legislatures and their own contracts by the time the federal report lands. Whatever the companies say under oath today will be on the record when those rules get written.
References
Committee of the Whole meeting details and agenda, Oct. 5, 2026 (New York City Council Legistar)
New York City Council subpoenas Musk's SpaceXAI over AI safety (The Next Web, Sept. 29, 2026)
Trump unveils his new Super Intelligence Force (TechCrunch, Oct. 4, 2026)
Trump launches Super Intelligence Force with Jay Clayton as AI czar (SiliconANGLE, Oct. 4, 2026)
AI benefits warrant accepting some risks: Sam Altman (Reuters via Neos Kosmos, Oct. 5, 2026)
Our data center commitment and the Built Together program (Amazon, Matt Garman, Oct. 2, 2026)
AWS drops non-disclosure agreements for data center projects (DCD, Oct. 2, 2026)
Amazon ends NDAs with local governments over data centers (WNDU, Oct. 2, 2026)
Read on the web: https://theainextwave.com/editions/2026-10-05/