TechNewsFirst

Malone News published a piece on September 19, 2026 arguing that the current wave of AI-safety alarm, a researcher quitting in fear, a company boss calling for slowing down, a bill to ban "artificial superintelligence," is not a genuine response to a genuine risk. The claim is that it is a regulatory-capture play: Anthropic, OpenAI, and Google can absorb the cost of compliance and litigation that a ban or a testing regime would impose; smaller labs and open-weight developers cannot. The piece draws a direct line to the 1906 Meat Inspection Act, arguing that a public panic over "The Jungle" let large meatpackers write their preferred rules into federal law and shrink the number of competing packing houses.

That is a real pattern with a name, regulatory capture, going back to George Stigler's 1971 work, and a real historical precedent behind it. The problem is what the piece asks readers to believe is happening right now, and which bill is supposedly doing the capturing.

What checks out

The chronology in the piece is mostly accurate against independent reporting. On September 3, 2026, Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act, which would permanently prohibit development of a broadly defined "superintelligent" AI, pause frontier development pending a new federal safety regulator, and impose corporate dissolution and up to 20 years in prison for violations (Sanders' Senate press release, September 3, 2026). Five days later, on September 8, Anthropic researcher Jacob Coxon posted on X that he had resigned, warning that "neither company is acting responsibly" and that both OpenAI and Anthropic were racing toward self-improving superintelligence (NBC News, September 2026; Time, September 15, 2026). Anthropic's own alignment science lead, Evan Hubinger, publicly agreed with him rather than distancing the company from the claim, writing that he personally believes there is more than a 10 percent chance AI kills everyone within a decade, an estimate, not a measured fact, and one made by an Anthropic employee about Anthropic's own field (CNBC, September 9, 2026). Four days later, Anthropic CEO Dario Amodei published "We Must Pace the Frontier," calling for the industry to slow capability gains by a year or two (darioamodei.com, September 12, 2026).

Anthropic's political spending is also real and well documented, independent of anything in the Malone piece. Anthropic gave Public First Action, a 501(c)(4) that backs candidates favoring AI safeguards, $20 million in February 2026, then doubled that to $40 million total in July 2026, citing the risk posed by its own increasingly capable models as the reason (Anthropic's own newsroom, February and July 2026; Axios, July 22, 2026).

Where the story breaks

Here is what the piece does not do well: it treats "the AI industry" as a single beneficiary of a single regulatory push, when the actual political money in 2026 runs mostly in the opposite direction. Leading the Future, the super PAC backed by OpenAI president Greg Brockman, Andreessen Horowitz, Palantir co-founder Joe Lonsdale, and Perplexity, has raised more than $140 million (NOTUS, 2026) largely to defeat candidates who favor disclosure and safety rules. It spent $7.6 million against New York Assembly member Alex Bores, author of New York's RAISE Act, a comparatively mild transparency law requiring frontier labs to disclose safety protocols (TechCrunch, 2026; The Nation, 2026). That is more than three times what Anthropic put into Public First Action for the entire cycle. If Big Tech is engineering a panic to write itself favorable rules, the largest pool of AI-industry political money in the country is doing the opposite: fighting the mildest version of those rules that has actually passed anywhere.

The bill Malone spends most of its argument on, the Sanders-Casar ban, is also a strange one to pin on Anthropic's interests specifically. Anthropic is mid-IPO. It raised $65 billion in May 2026 at a $965 billion valuation and confidentially filed for a public offering the following month, with reporting pointing to a targeted valuation near $2 trillion by October, a figure that remains speculative until pricing is actually set (Fortune, June 1, 2026; further reporting aggregated by Graniteshares citing Bloomberg, 2026). That valuation is built on the premise that Anthropic keeps scaling. A bill that bans and pauses frontier development does not protect that bet; it guts it. A testing-and-disclosure regime that only the biggest labs can staff, the kind of policy Anthropic's own PAC money and OpenAI's own September 9 essay from Chris Lehane actually argue for, is a coherent regulatory-capture story (OpenAI, September 9, 2026, "The AI policy window is open"). A full ban on the thing a $2 trillion valuation depends on is not. The piece never separates these two different asks, a ban bill with no obvious industry beneficiary and a disclosure-and-testing regime that plausibly has one, and that is the load-bearing gap in its argument.

The sourcing problem

The financial detail in the piece, Anthropic's PAC donations, Amodei's own political giving, named employee donations, is attributed to "Sayer Ji's analysis," dated September 13, 2026. Independent of whether Ji's specific figures are right, Ji is the founder of GreenMedInfo, a site Media Bias/Fact Check rates for pseudoscience and low factual reporting, and Ji was named in the Center for Countering Digital Hate's 2021 "Disinformation Dozen" report for spreading vaccine misinformation, including a false claim that Pfizer's COVID vaccine killed more people than the virus itself (Media Bias/Fact Check; Wikipedia, "Sayer Ji"). Ji disputes that framing, and he has a real point buried in the dispute: Meta later found the CCDH's headline statistic, that the twelve named accounts drove 73 percent of vaccine misinformation, was off by roughly three orders of magnitude, closer to 0.05 percent of views, a correction worth stating plainly rather than pretending it doesn't exist. It does not erase the separately documented false claims Ji made during the pandemic. None of this makes the underlying $20 million and $40 million Anthropic donation figures wrong; those numbers are independently confirmed by Anthropic's own press releases and by Axios, CNBC, and TechCrunch reporting with no connection to Ji at all. But routing a financial-interest argument through a source with Ji's specific track record, when better sourcing was sitting in the open, hands a skeptical reader an easy reason to dismiss a claim that happens to be true.

The part the piece treats as more solid than it is

The Coxon resignation is placed, by its position inside a "coordinated panic" timeline, as part of the machinery. Every independent account of it, NBC News, Time, CNBC, Scientific American, describes it the same way: an unplanned personal post that went viral by accident, not a scheduled announcement. Coxon has said he hit send with no expectations. Hubinger's response reads as a researcher agreeing with a colleague's fear, not a company deploying a talking point; publicly agreeing with an alarming claim about your own employer's technology is a strange thing to fake for PR purposes, since it invites exactly the kind of scrutiny Anthropic is now getting. There may be a case that Anthropic benefits from the reaction to Coxon's post regardless of whether it orchestrated it. There is no public evidence, in the Malone piece or found independently for this article, that the resignation itself was choreographed.

What would make the capture story work

The regulatory-capture pattern from 1906 is real, and the actors who benefit from a rulebook only they can afford to comply with are worth naming. But the piece to write is about the disclosure-and-testing regimes that Anthropic's PAC money and OpenAI's own September policy shift are actually pushing for, where a credible case exists that compliance costs favor whoever already has the compliance department built. The bill Malone built its headline around, a bill that would ban the product a $965 billion company is about to take public, is the wrong exhibit for that argument.