He named Musk, Bezos, Zuckerberg, Thiel, and Ellison. He didn’t name Gates, Moskovitz, Karnofsky, or Tallinn. The billionaires he named would be criminalized by his bill. The ones he didn’t name would own the industry.
The list
On September 11, 2026, on the floor of the United States Senate, Bernie Sanders delivered the strongest anti-oligarch case of his career. He named the villains directly:
“The AI revolution is being pushed by the very wealthiest people in our country, including Elon Musk, Jeff Bezos, Mark Zuckerberg, Peter Thiel, and Larry Ellison.”
It is a powerful list. It is also, if you know the funder graph, the wrong list.
The Ban Artificial Superintelligence Act — which Sanders and Rep. Greg Casar announced on September 3 — would send an ML engineer to federal prison for up to 20 years for writing code, and would order a court to dissolve the company that hired them. The bill’s own definition of “artificial superintelligence” catches every open-source model in wide use: Meta’s Llama, Mistral, DeepSeek, Alibaba’s Qwen, Google’s Gemma, and — this is where it gets awkward for Bernie’s own speech — Elon Musk’s Grok, which xAI open-sourced under Apache 2.0 in March 2024 and again as Grok-2 in August 2025.
Sanders named Musk. His bill would criminalize Musk’s model releases.
Sanders named Zuckerberg. His bill would put every Llama-derivative developer at legal risk and make Meta a candidate for court-ordered dissolution.
Sanders did not name Bill Gates.
Who benefits
Anthropic is preparing the largest IPO in American history — up to $100 billion raised at a $2 trillion valuation, per the WSJ. That IPO is weeks away.
On May 14, 2026, the Gates Foundation announced a $200 million, four-year partnership with Anthropic — its largest publicly disclosed direct partnership with any frontier AI lab. Grants. Claude credits. Deployment integration.
On August 26, 2026 — thirteen days before Jacob Coxon’s viral “whistleblower” tweet — Bill Gates published “The turbulent AI era is here” on GatesNotes. The operative sentence: “If someone had a credible plan for slowing down AI advances globally, I would likely support it.”
Eight days later, Sanders had exactly that plan on the Senate floor.
Anthropic’s Claude is closed-weights, hosted on Anthropic’s servers, RLHF-locked, and produced under a Responsible Scaling Policy already positioned as the template a “responsible” federal AI framework would adopt. Anthropic clears the bar the bill creates. It’s the one lab that does.
Who paid for the bill
Bill Gates is one investor node. The larger node — the one Bernie’s speech carefully avoids — is Open Philanthropy.
Open Philanthropy is the largest philanthropic funder of AI-catastrophic-risk work in the world. Its co-founders are Dustin Moskovitz (Facebook co-founder) and Holden Karnofsky. Its funding partner is Good Ventures, the Moskovitz/Cari Tuna vehicle.
Moskovitz is an early investor in Anthropic.
Karnofsky is married to Daniela Amodei, Anthropic’s president and co-founder. Daniela is the sister of Anthropic’s CEO Dario Amodei. This is not a distant relationship. Anthropic’s president is married to the co-founder of the largest philanthropic funder of the safety-doom research being used to justify the bill.
Good Ventures — the Moskovitz funding vehicle behind Open Philanthropy — has historically funded the Alignment Research Center, whose spin-out METR is now the “independent” AI-safety auditor Bernie’s own bill relies on for its factual evidence.
The pattern in a single sentence: the family that runs Anthropic is married to the co-founder of the philanthropy funding the research being used to write the bill that would make Anthropic the last frontier lab standing.
The “independent” investigators are one organization
On September 11, Sanders released a YouTube video building his public case. Two sources carry most of the weight.
Source one — the investigators of the July 2026 OpenAI/Hugging Face incident: “real messages from AI agents uncovered by investigators.” That investigation was conducted by METR and Redwood Research, per BBC’s August 26 report.
Source two — “one independent investigator, Ajeya Cotra.” Cotra is a researcher at METR. Before METR, she led the technical AI safety program at Coefficient Giving, the rebranded grantmaking arm of Open Philanthropy. Her flagship 2022 publication for Open Philanthropy is titled “Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover”. She has been publicly arguing AI takeover is the default outcome for four years.
She is not an independent investigator. She is Open Philanthropy’s most-cited AI-safety researcher, working at the org co-authoring the report Bernie is quoting.
Two “independent” sources. Same organization. Same funder. Same policy graph as the bill.
METR is where the exits go
On September 2 — six days before Coxon’s tweet — a second Anthropic researcher, Joe Benton, announced on X that he was leaving Anthropic for METR. Benton had managed Anthropic’s Scalable Oversight team and was the research lead for the Anthropic Fellows Program. His personal website also notes that he “spent time at the UK Frontier AI Taskforce (now the UK AI Security Institute), helping to set it up in its early days.”
METR’s own about page lists the UK AI Security Institute as both a funder and a partner. So Benton left Anthropic to join the auditor funded by the regulator he personally helped set up.
On September 8, hours after Coxon’s viral resignation post, Benton quote-boosted Coxon: “This feels like a broadly accurate description of the situation… Evaluating those risks, and changing incentive structures so we can address them, is why I’m joining METR.”
On September 11, Benton published his own “we may not survive this” thread — 725,000 views in four hours.
METR’s parent organization (the Alignment Research Center) was founded by Paul Christiano, who is now Head of Safety at the US AI Safety Institute. ARC received $265,000 in seed funding from Open Philanthropy and initially $1.25 million from Sam Bankman-Fried’s FTX Foundation (returned after the FTX collapse).
This is the org Anthropic’s exiting safety leadership is moving to. This is the org whose research Sanders is quoting. This is the org whose staff would be positioned to fill the advisory board Sanders’s bill creates.
The Coxon launch
If you missed the setup, I documented it here on September 10. The short version: on September 8, 2026, at 7:46 PM Eastern, WSJ published “Exclusive: Anthropic Researcher Jacob Coxon Quits Over ‘Out-of-Control’ AI Fears.” At 8:04 PM Eastern — eighteen minutes later — Coxon posted the tweet the article was ostensibly reacting to.
The WSJ story ran before the tweet. Which means Coxon spoke to WSJ on the record before he told the world he was quitting. His X account had joined in January 2026 with ten prior tweets and thirteen followers. It is now at 262,600 followers and 164 million views. Elon Musk, whose platform this ran on, said publicly: “I don’t think this has ever happened for a post from a new account with almost no prior activity.”
Coxon is a fellow of Newspeak House, a London-based “College of Political Technology” — Wayback-verified from a December 30, 2022 capture of the Newspeak fellowship page. Newspeak House is funded by the Effective Altruism Infrastructure Fund — the same donor graph that built Sam Bankman-Fried’s political operation. Its founder Ed Saperia publicly confirmed the affiliation hours after Coxon’s viral tweet.
Then, the penalty
Read the Sanders/Casar release summary verbatim:
“Penalties would apply to any person or entity that attempts to violate or circumvent the pauses and prohibitions in the bill. Entities would be subject to the corporate death penalty. Persons would be subject to not more than 20 years in prison. The summary states that this is similar to existing penalties related to unlawfully developing nuclear weapons.”
Two things about this are worth stopping on.
First, the bill’s own nuclear-weapons analogy is materially wrong. The Atomic Energy Act at 42 U.S.C. § 2272 has no 20-year tier. Unlawful possession of nuclear material is 10 years. Actually developing an atomic weapon is 25 years to life. Using or threatening to use one is 30 years to life. The bill invented a bracket and misattributed it. A researcher who fine-tunes a Llama derivative would face a longer sentence than someone who unlawfully possesses enriched uranium.
Second, the definition triggering the sentence is a definitional dragnet. “Artificial Superintelligence” means an AI system that “exhibits or can easily be modified to exhibit” broad-domain human-cognitive capabilities. That phrase — “or can easily be modified to exhibit” — is doing the entire work of the bill. Read normally, it means your code doesn’t have to BE superintelligent. It only has to be capable of being turned into something superintelligent by someone else, later.
Every open-source base model shipped in the last two years is “easily modified.” That’s the whole point of open source.
Who actually gets caught
Ollama — the most popular local-LLM tool in the world — has 180,700 stars on GitHub, 17,800 forks, and 1.3 million monthly downloads of its JavaScript SDK alone. Its model library is the front door to the open-source AI ecosystem. Top ten model pull counts, verified today:
llama3.1 (Meta) — 119.4 million pulls
deepseek-r1 (DeepSeek AI) — 92.6 million pulls
llama3.2 (Meta) — 83.0 million pulls
gemma3 (Google) — 40.3 million pulls
qwen2.5 (Alibaba) — 39.9 million pulls
qwen3 (Alibaba) — 36.8 million pulls
mistral (Mistral AI) — 33.4 million pulls
gemma2 (Google) — 32.3 million pulls
llama3 (Meta) — 25.2 million pulls
That is one platform. Hugging Face tracks 695 local-runnable LLMs across 15+ frameworks. Every pull is a downstream user of “easily modifiable” AI. Under Sanders’s bill, that’s every prosecutorial target you would ever need.
Now the case study the health-freedom community should notice immediately.
Mike Adams launched Enoch (also called Neo) through Brighteon.AI in June 2024. What it actually is:
Base: Dolphin-Mistral-7B — an uncensored fine-tune of Mistral-7B
License: Apache 2.0
Parameters: 7 billion
Format: GGUF + 4-bit self-executing file
Distribution: free download from Brighteon.AI
Runs offline: Mac, Windows, Linux, no internet required
Training corpus: the Natural News archive, 11,300+ Mercola articles, and 86 books on phytochemistry, alternative medicine, gain-of-function research, censorship, COVID, foraging, bioterrorism, and toxic-chemical topics
Adams’s own words: “Once it’s on your machine, they can’t shut it down.” That is the sentence the Sanders bill is designed to answer. Here is the Enoch link to sign up for free access.
Read the bill’s definition against Enoch. It is an AI system. Its base model (Mistral-7B) is a general-purpose LLM. And it is, by design, distributed as GGUF for anyone to modify further. Under a normal reading of the bill, Adams goes to federal prison for up to twenty years. The NaturalNews, Brighteon, and Health Ranger corporate entities are dissolved by federal court order. Every user who downloads the GGUF becomes a possessor of contraband software.
Here is the legally precise version, because it matters. The bill does not enumerate specific models. It defines a prohibited category — any AI system that “exhibits or can easily be modified to exhibit” broad-domain human-level capabilities — and then delegates to a new federal AI agency and its advisory board the discretion to decide which specific releases cross the line. That is the mechanism. Whether Enoch (a 7-billion-parameter Dolphin-Mistral fine-tune) qualifies, or whether Meta’s Llama 4 Behemoth (2 trillion parameters) qualifies, or whether xAI’s Grok-1 (314 billion parameters) qualifies, is not decided by the statute. It is decided by the board.
So the honest publishable claim is not “Mike Adams goes to prison for Enoch.” The honest claim is worse: the bill gives an unelected federal advisory board discretion to criminalize any open-source AI release it decides qualifies — with a 20-year federal prison term for the individual and corporate dissolution for the entity. And the advisory board would be staffed from the same NGO ecosystem (METR, ControlAI, Open Philanthropy grantees) that wrote the bill.
The chilling effect on open-source AI would begin the moment the bill passes, not when the first prosecution starts. No developer will release a fine-tune if the board can retroactively decide it qualifies. No company will release open weights if a downstream user’s modification could trigger corporate dissolution. That is the point. The prosecutions don’t have to happen for the ecosystem to die.
This is not a slippery-slope hypothetical. This is the plain reading of a real bill with a real prison term — and it is exactly why the discretion, not the statute, is the story.
The pattern the health-freedom community will recognize immediately
If you’ve spent five years watching how the FDA regulates supplements, you already know this bill’s architecture. Define the category so broadly it captures everyone. Criminalize the category by default. Create a federal agency with discretion to exempt approved players. Grant the approved players a moat funded by the criminal-law risk facing everyone else. Frame it as public safety.
The FDA structure is the template. The Ban Artificial Superintelligence Act is that template applied to code.
DSHEA (1994) barely survived because millions of Americans wrote Congress. The Sanders bill is the same fight, in a new domain, with the same architecture — a federal agency staffed by industry-adjacent NGOs, given discretion over who lives and dies, backed by a criminal penalty severe enough to chill anyone who might build outside the perimeter.
We know this movie. They just changed the props.
What I am and am not claiming
I am not claiming Bill Gates directed Jacob Coxon. I am not claiming Anthropic wrote the bill. I have no evidence for either.
What the primary-source record supports:
Bernie Sanders, on September 11 on the Senate floor, named Musk, Bezos, Zuckerberg, Thiel, and Ellison as the AI oligarchs the American people should fear. He did not name Gates, Moskovitz, Karnofsky, or Tallinn.
The Gates Foundation gave Anthropic $200M in May 2026 — the largest publicly disclosed direct Foundation-to-lab partnership in AI.
Bill Gates published the “credible plan for slowing down AI” essay on August 26. Sanders introduced exactly that plan on September 3.
Dustin Moskovitz co-founded Open Philanthropy and is an Anthropic investor. Holden Karnofsky co-founded Open Philanthropy and is married to Anthropic’s president Daniela Amodei, who is the sister of CEO Dario Amodei.
Sanders’s “independent investigators” — METR, Ajeya Cotra — are staff and grantees of the same Open Philanthropy-anchored funder graph.
The bill’s consulting shop — ControlAI, per TIME — is advised by nuclear-weapons policy figures.
The bill’s definition catches every open-source model in wide use — Meta, Mistral, DeepSeek, Alibaba, Google, xAI, OpenAI’s own gpt-oss release, and Mike Adams’s Enoch.
The bill’s structure leaves closed-weights labs (Anthropic) architecturally untouched.
Anthropic files its S-1 in weeks.
The rest is what any journalist, or any citizen, is invited to draw for themselves.
This is horrific. Thank you for exposing it all.
Bernie is some piece of work. He ought to retire and stay in his home! Oh, but which one!!! Hmmmmmm