Monday, 5 October 2026
Four Came, One Was Subpoenaed
All Fifty-One Members of New York’s City Council Sit Down Today to Question Four AI Companies Under Oath
The Council convenes as a Committee of the Whole this morning — every one of its fifty-one members in the room at once, a format it keeps for the things it does not want delegated — to take testimony on the package of AI bills Speaker Julie Menin unveiled on 25 September. Fortune’s account of that package is the part worth reading closely: any AI system marketed, sold or deployed in the city would have to pass outside validation for data quality, bias, privacy and security against standards set by the Office of Cyber Command; every system would need a kill switch a human can actually reach; New Yorkers could sue an AI company when somebody else jailbreaks its product and the harm was foreseeable; and whistleblowers would take a cut of the fines, which run to $25,000 per instance and attach to the validator as well as the business deploying the thing. OpenAI, Anthropic, Google and Meta are sending representatives. Three of those four, by Route Fifty’s reporting, agreed only after the Council said out loud that it had subpoena power. Elon Musk’s SpaceXAI never answered the invitation at all and was subpoenaed on 28 September — the first time Menin has used that power since becoming Speaker — with the Council saying it will go to state court if the company still does not appear. “These companies have a significant presence in New York City and collectively employ tens of thousands of New Yorkers,” Menin said. “Given the high stakes, these firms owe it to their workers and, above all, to the public to come before the Council, answer our questions, and provide input on our proposed legislation under oath.” Washington has passed no AI safety law. A city council has subpoena power and a room that holds fifty-one.
Source: Route Fifty, September 29, 2026
Altman Says the World Should Accept Some Bad Things Happening, and Names What He Won’t Accept
Speaking to POLITICO’s Brendan Bordelon for the Decoded newsletter, published Sunday, Sam Altman put the trade-off in the plainest terms he has yet used: “we believe that the world should accept some bad things happening for the benefits of this technology.” He drew the line immediately afterwards — not at hacks or misuse, which he said he would not trade away broad access to prevent, but at “the really catastrophic risks,” naming “a serious loss of control to AI” among them. The interesting half is what he said about Anthropic. Asked to distinguish the two companies, Altman described the opposing worldview as the belief that the technology “is going to get so powerful, and it’s so dangerous, that a single lab in San Francisco should have it and make sure nothing bad happens, and kind of figure out how to dole out the benefits” — a position he called “a completely unacceptable trade-off,” and a framing that lands close to the regulatory-capture charge Trump-aligned tech advisers have been aiming at Anthropic for months. This is the same Altman who last week agreed with Dario Amodei that development of the most advanced models should slow down, and whose company has lately endorsed third-party evaluations of the kind New York wants to mandate. The gap between the two labs is narrowing on what should be checked. It is not narrowing at all on who gets to hold the thing.
The AI Force Trump Announced in September Now Has Four Names, and One of Them Is Investigating Two of the Companies It Will Coordinate With
When the president said on 19 September that he would stand up an “AI Force” with a czar at the top of it, there were no names, no reporting line and no mandate — we covered the announcement of intent at the time, and that was all there was. On Sunday, hours after a White House meeting with leaders from OpenAI, Anthropic, Google, Meta, Nvidia and SpaceX’s AI division, he filled it in on Truth Social: the Super Intelligence Force, led by Director of National Intelligence Jay Clayton, with FTC Chairman Andrew Ferguson, Pentagon Under Secretary for Research and Engineering Emil Michael and OPM Director Scott Kupor beside him, all four reporting to Trump and chief of staff Susie Wiles. “Following the Historic White House Accord on Super Intelligence, where the World’s most important Technology Companies came together, and confirmed their responsibilities to the American People, I am announcing the formation of the Super Intelligence Force (SIF),” he wrote. Its job, as described, is engagement rather than enforcement: consumers, public-interest groups, religious organisations, infrastructure providers and the companies themselves. Note the second name on that list. Ferguson’s FTC opened a formal consumer-protection investigation into OpenAI, Anthropic and their safety auditor on 30 September, and it is his own line about liability — the person who wields the hammer faces the consequences, not the hammer — that the industry has been quoting back all week. He now sits on a body whose stated purpose is coordinating with the people holding the hammer.
Source: CBS News, October 4, 2026
On Friday, Google Takes Its Good Models Away From Everyone Who Isn’t Paying Twenty Dollars
From 9 October, Gemini’s free tier gets Flash-Lite and nothing else — no Flash, no Pro. The $4.99-a-month AI Plus tier keeps Flash and loses Pro. Pro access now begins at AI Pro, $19.99 a month, which in exchange picks up Deep Think, the parallel-reasoning mode that until now belonged to the $99.99 and $199.99 tiers. Google is also adding low, medium and high effort levels, described in its own support documentation in terms that give the whole thing away: raising the effort “will increase the model’s ability to complete tasks and provide more thorough answers to queries, but will also use more of your limit.” The timing is the story. Gemini 4 Argon shipped last Wednesday, to considerable noise about frontier capability, and the first consumer-facing consequence of that launch is that the people who were using Google’s second-best model for nothing now get its weakest one. Frontier models are expensive to serve. Somebody was always going to be asked to stop using them, and it is never going to be the $199.99 tier.
Source: 9to5Google, October 3, 2026