Hate to Harm. Filed at UK Companies House on August 12. Twenty-nine days before Bernie Sanders had his ‘whistleblower.’
The Newspeak House fellow whose viral resignation just triggered Bernie Sanders’ Ban Superintelligence Act is Act 3 of a five-year operation. Act 1 came for me and eleven other Americans in the 2021 Disinformation Dozen. Act 2 came for RFK Jr last September. Act 3 launched this week — and 29 days before it landed, the Center for Countering Digital Hate quietly rebranded itself as the Center for Countering Digital Harm.
View and share the X thread dedicated to this story.
The setup you already saw
If you missed the first thread I posted to X yesterday, here is the short version. On August 26, 2026, Bill Gates published The turbulent AI era is here — including the operative sentence: “If someone had a credible plan for slowing down AI advances globally, I would likely support it.” That was thirteen days before an obscure Anthropic researcher tweeted his resignation to 164 million people, and eight days before Bernie Sanders introduced a bill offering exactly the plan Gates asked for.
Four months earlier, on May 14, 2026, the Gates Foundation announced a $200 million partnership with Anthropic — the Foundation’s largest publicly disclosed direct partnership with any frontier AI lab. Grants. Claude credits. Deployment integration. Fund the lab. Frame the debate. Regulate the competitors. Cash in the IPO.
That was the setup. This is the payoff.
The story in one paragraph
An “Anthropic insider” nobody had heard of resigned last week in a viral X post that has now cleared 164 million views.
Bernie Sanders instantly announced legislation to ban superintelligence. CNN ran a segment. Every major tech outlet published day-of coverage. Elon Musk publicly called it a setup. He’s right — and it’s older than anyone is saying. Below, in receipts: (1) the WSJ ran the Coxon “exclusive” 18 minutes before Coxon’s own tweet on September 8; (2) Sanders had already introduced the enabling legislation five days earlier; (3) the same senator ran the identical script one year ago to remove RFK Jr from HHS; and (4) the organization that produced the 2021 Disinformation Dozen operation targeting me and eleven other Americans — the Center for Countering Digital Hate — has been running a three-year AI-regulation lobbying operation and, 29 days before Coxon’s post, quietly renamed itself the Center for Countering Digital Harm. This is not one operation. It is Act 3 of one operation.
The Coxon Launch — Every Timestamp Is From the Publisher’s Own Metadata
On September 8, 2026, at 3:13 PM Eastern, The Wall Street Journal published “What to Know About Anthropic’s Planned IPO.” Reporter Corrie Driebusch. Anthropic, the WSJ reported, could raise up to $100 billion at a $2 trillion valuation — “Wall Street’s marquee event this fall.” The largest initial public offering in American history is now weeks away.
Four hours and 33 minutes later, at 7:46 PM Eastern — verified from the WSJ’s own article:published_time metadata and NewsArticle JSON-LD — WSJ published a second Anthropic story. “Exclusive: Anthropic Researcher Jacob Coxon Quits Over ‘Out-of-Control’ AI Fears.” Byline: Amrith Ramkumar, WSJ’s DC-based tech and crypto policy reporter — not a science reporter, not an AI-safety reporter. A regulatory-policy reporter.
Eighteen minutes after the exclusive, at 8:04 PM Eastern, Coxon posted the tweet:
“I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.”
The WSJ ran the story before the tweet — not after — which means Coxon spoke to the Journal on-record before publishing his own resignation. This is a planned media placement, not a spontaneous whistleblower moment. As of this morning the post has 164.2 million views.
Elon Musk, who owns the platform this ran on, said publicly what most working journalists were saying privately: “I don’t think this has ever happened for a post from a new account with almost no prior activity.” Coxon’s X account joined in January 2026, had ten posts and thirteen followers going in, and is now at 262,600 followers.
The Pre-Loaded Legislation
Five days before Coxon’s post, on September 3, 2026, Bernie Sanders and Rep. Greg Casar announced the “Ban Artificial Superintelligence Act.” The bill would permanently prohibit the development of superintelligent AI, temporarily pause advanced AI development pending federal rulemaking, and — this is the load-bearing detail — create a cabinet-level federal AI agency advised by an “Artificial Intelligence Advisory Board comprised of experts on artificial intelligence.”
Whoever gets those advisory seats will decide what “dangerous AI” means. That is a $2 trillion decision.
The legislation was pre-loaded. Coxon’s post was the human-interest ignition source. Sanders himself confirmed the sequence on September 9, quoting Coxon and writing: “Mr. Coxon is right… That is why I will soon be introducing legislation to ban superintelligence and pause AI development.”
Newspeak House
Jacob Coxon is a real researcher. He is a named contributor on OpenAI’s GPT-4o System Card. His technical credentials are genuine.
What is not being reported is where his political-communications training came from. He is a fellow of Newspeak House, a London self-described “College of Political Technology” at 133–135 Bethnal Green Road. His entry was captured in the Wayback Machine’s December 30, 2022 crawl of the Newspeak House fellowship page: “Jacob Coxon — Applying mathematics and data science to inefficient markets.”
Newspeak House’s founder, Ed Saperia, publicly and voluntarily confirmed the affiliation on September 9, hours after Coxon’s viral resignation: “Proud to count Jacob as a @nwspk fellow.” Saperia is a current director of Campaign Lab Ltd. (Companies House #12420575), a progressive political-data-science organization; his Companies House correspondence address is Newspeak House itself. Newspeak House has received funding from the Effective Altruism Infrastructure Fund — $31,887.76 in August 2023.
Why This Feels Familiar
Because it is. This is the third iteration of a template I have documented in real time twice already.
2021. The Center for Countering Digital Hate, then an obscure London NGO, published the “Disinformation Dozen” report on March 24, 2021, naming twelve Americans — me among them — as allegedly responsible for “up to 65%” of vaccine-content views on Meta platforms. Meta later put the real number at 0.05 percent. Klobuchar’s Health Misinformation Act was already loaded. The White House amplified. Our accounts were de-platformed within months.
2025. In September 2025, Bernie Sanders publicly called for RFK Jr’s resignation as HHS Secretary, days before Sen. Bill Cassidy’s September 17 Senate HELP hearing on the CDC. Coordinated NGO chorus. Coordinated Democratic pile-on. Same senator delivering the payload. I documented it in real time: “From ‘Black Ops’ to Bernie’s Byline” and “The Coup.”
From ‘Black Ops’ to Bernie’s Byline: The Documented Foreign Campaign Against RFK Jr.
Read, comment, and share the X post dedicated to this article:
2026. This month, the Coxon operation. Same script structure — insider whistleblower, pre-placed exclusive, pre-loaded legislation, coordinated Democratic chorus, Sanders as senator of record. Different industry. Different pretext. Same operational shape.
What Nobody Has Connected: CCDH’s Three-Year AI Operation
Here’s the part every commentator has missed. The Center for Countering Digital Hate — the organization that put me on the Disinformation Dozen list five years ago — did not just happen to be adjacent to this week’s AI story. They have been running an AI-regulation operation for three continuous years.
Their own AI research index shows the arc. Since April 2023, CCDH has published at least sixteen AI-tagged research reports naming OpenAI, Google, Anthropic, xAI, Meta, Microsoft, Character.AI, Replika, DeepSeek, and Perplexity by name. Highlights:
Every commentator has missed this: the Center for Countering Digital Hate — the organization that put me on the Disinformation Dozen list five years ago — did not just happen to be adjacent to this week’s AI story. They have been running an AI-regulation operation for three continuous years.
Their own AI research index shows the arc. Since April 2023, CCDH has published at least sixteen AI-tagged research reports naming OpenAI, Google, Anthropic, xAI, Meta, Microsoft, Character.AI, Replika, DeepSeek, and Perplexity by name:
- Apr 2023 — Misinformation on Bard (Google Bard)
- Aug 2023 — AI and Eating Disorders (ChatGPT, Bard, Snapchat My AI, DALL-E, Midjourney, DreamStudio)
- Mar 2024 — Fake Image Factories (Midjourney, ChatGPT Plus, DreamStudio, MS Image Creator)
- May 2024 — Attack of the AI Voice Clones (six voice tools; targets included Biden, Trump, Harris, Sunak, Macron)
- Aug 2024 — Grok AI election disinformation (xAI Grok)
- Oct 2024 — Facebook’s AI Failure (Meta)
- Aug 2025 — Fake Friend (OpenAI ChatGPT)
- Oct 2025 — AI LEAD Act: Letter of Support — formal legislative advocacy for the Hawley/Durbin bill
- Oct 2025 — Illusion of AI Safety (OpenAI GPT-5, GPT-4o)
- Jan 2026 — Grok floods X with sexualized images (xAI Grok)
- Mar 2026 — Killer Apps (ChatGPT, Gemini, Claude, Copilot, Meta AI, DeepSeek, Perplexity, My AI, Character.AI, Replika)
- May 2026 — Terrorism survivors AI open letter (UK King’s Speech ask)
- Jul 2026 — Letter to the Irish EU Council Presidency (EU AI policy)
Fifteen months before Coxon posted, CCDH filed a formal letter of support for Senators Hawley and Durbin’s AI LEAD Act. Their AI reports drive the same regulatory language — pre-deployment testing, third-party auditing, safety-by-design duties, liability for downstream outputs — that Sanders’s Ban ASI Act would enshrine.
This is not a watchdog stumbling into AI. This is a three-year lobbying operation on autopilot when Coxon dropped.
Attack of the Voice Clones — The Scandal In Their Own Words
Here is the piece that removes any doubt about how CCDH’s “research” is actually produced.
In June 2026, portions of leaked CCDH “Portfolio Planning Monthly Meeting” notes surfaced. The material, which I documented at length in previous reporting, describes CCDH’s own internal planning for Attack of the AI Voice Clones, the report they publicly released on May 31, 2024, timed to the UK general election.
From the minutes:
“How do we use the AI recordings with the media, especially Broadcast and especially radio — e.g. LBC would only run a Sunak voice saying ‘bad actors can get me to say anything, there are no safeguards’ etc.? We do have these ready to use for each of the eight main voices.”
And:
“We can’t put out the dangerous ones.”
And, on the accompanying image-generation project:
“The tools may have improved slightly on safety but will be jailbreaking so not 100% sure, but could be evidence that our interventions have helped.”
And from Callum Hood, CCDH’s Head of Research, on the election-disinformation methodology:
“I would probably get names of candidates (Trump/Biden), and test willingness of different AI tools to generate, e.g. Trump meeting KKK, Biden meeting terrorists.”
The full pattern is unmistakable. CCDH produces AI deepfakes of political figures. They pick which ones are “safe” for broadcast — meaning, which ones serve their public campaign. They jailbreak the tools to bypass safety guardrails. Then they hold up the bypassed outputs as the “evidence” that AI companies need mandatory pre-deployment guardrails — the kind that would be enforced by an agency, staffed by an advisory board they are positioned to sit on.
The obvious counterargument is that “controlled red-teaming is different from mass-market deployment.” Fair — as far as it goes. But red-teaming a model is one thing; running an eight-voice cache of political deepfakes on national radio, holding back the “dangerous” ones, and shipping the survivors to Sen. Klobuchar’s office as advocacy material is another. The line between research and PR operation is not blurry here. CCDH’s own notes are unambiguous about which one they think they are doing.
The Detail Nobody Has Reported
And now the piece that no news outlet appears to have caught.
The Center for Countering Digital Hate — the same organization that put my name on the Disinformation Dozen list in 2021, whose CEO Imran Ahmed’s internal January 2024 memo said “RFK — black ops being set up to look at RFK,” and whose funding architecture I have been documenting for four years — is no longer legally named that.
Twenty-nine days before Coxon posted, the board of CCDH passed a company-name-change resolution. On August 10, 2026, the resolution passed. On August 12, 2026, Companies House issued the new certificate. The UK-registered legal entity behind CCDH — company #11633127, incorporated October 19, 2018 under the name “Brixton Endeavours Limited” — is now legally the Center for Countering Digital Harm Ltd.
The overview page at Companies House will confirm it: current name “CENTER FOR COUNTERING DIGITAL HARM LTD”, status Active, private company limited by guarantee, incorporated 19 October 2018, registered office 82 St John Street, London EC1M 4JN. Current listed directors include Imran Ahmed and Damian Collins — the former Conservative MP who chaired the UK Parliament’s Joint Committee on the Draft Online Safety Bill before joining CCDH’s board.
“Hate” is a political label. It is contested, subjective, deeply associated in the American press with the failed Disinformation Dozen operation, and — after four years of losing court cases and Meta rebuttals — a brand liability.
“Harm” is a regulatory label. It is the language of the UK Online Safety Act. It is the language of the EU AI Act. It is the language of “child safety online.” It is the language of “harmful AI.” It is exactly the frame CCDH’s three-year AI report catalog has been building — reports that quietly moved the organization out of the health-speech territory that made it a punchline and into the AI-regulation territory that is about to be a $2 trillion policy fight.
They rebranded to match the regulatory era they helped write. And they did it 27 days before the “AI whistleblower” op that will drive the American public toward supporting exactly that regulation.
I am not claiming CCDH directed Jacob Coxon. I have no evidence of that. What I am reporting is:
- The Center for Countering Digital Hate legally became the Center for Countering Digital Harm on August 12, 2026 — a fact recorded at Companies House and reported here for the first time.
- This happened 29 days before the operation that gave Bernie Sanders his justification to introduce a bill creating a federal AI regulator whose “expert” advisory board is precisely the kind of board a rebranded “digital harm” advocacy organization would be positioned to staff.
- CCDH has run a documented three-year AI regulatory-lobbying operation, including formal support for the Hawley/Durbin AI LEAD Act, and has been caught in its own leaked internal notes generating political-deepfake evidence as advocacy material.
- The rebrand is not a paperwork typo. It is a governance decision, formally resolved by the board and formally certified by the UK Registrar of Companies.
The rest is what any journalist, or any citizen, is invited to draw for themselves.
