Massive Florida OpenAI Lawsuit Reveals a Second Story: CCDH’s Private Playbook Looks Nothing Like Its Public Face
The group warning about AI deepfakes was producing its own.
The same group cast as an AI-safety watchdog in Florida’s blockbuster case against OpenAI was privately discussing how to “Kill Musk’s Twitter,” trigger regulation across the U.K., EU, and U.S., jailbreak AI tools, and weaponize synthetic media for policy and press advantage.
Share the X thread dedicated to this article.
Story at a glance
Florida’s complaint against OpenAI leans on CCDH’s research to portray the group as a credible authority on AI harms, especially harms to minors.
CCDH’s own internal planning memo shows a recurring strategic objective to “Kill Musk’s Twitter,” alongside plans to “Trigger EU & UK regulatory action” and drive “Progress towards change in USA and support for STAR.”
The same memo shows CCDH planning AI voice-clone and AI-image campaigns aimed at policymakers, broadcasters, and major newspapers.
CCDH explicitly discussed jailbreaking AI tools to generate political images and then using those outputs as evidence that AI companies had failed to install proper guardrails.
Internal discussions framed election-disinformation and child-safety content not just as public-interest work, but also as vehicles for “planting a flag,” driving profile growth, increasing SEO traffic, and supporting fundraising.
The resulting picture is not of a neutral watchdog, but of a sophisticated political-pressure operation whose private methods appear to run directly against its public posture.
A massive lawsuit filed by the State of Florida against OpenAI and Sam Altman is being presented as a landmark case about the harms of artificial intelligence: manipulated minors, addictive chatbot design, unsafe medical advice, and the commercial deployment of a system the state says was marketed as safe and trustworthy when it was not. That story alone is explosive.
But buried inside the same legal and documentary universe is a second story that may prove just as important: one of the lawsuit’s most useful outside validators, the Center for Countering Digital Hate (CCDH), whose antics I have been reporting on for the past 5 years, appears in its own internal planning documents as something very different from the sober, neutral, public-interest watchdog it presents to the world.
Florida’s complaint relies in part on CCDH’s research to support the claim that ChatGPT can encourage self-harm, provide dangerous guidance to teens, and operate without meaningful guardrails. In the complaint, CCDH is treated as a credible authority on what unsafe AI looks like and why inadequate controls matter.
Yet CCDH’s own internal “Portfolio Planning Monthly Meeting” memo reveals a private operating posture that is not merely activist, but overtly strategic, adversarial, and at points startlingly deceptive in relation to its public face. Over and over, across multiple 2024 entries, CCDH lists as a standing priority: “Kill Musk’s Twitter.” In the same recurring priority blocks, it lists “Advertising focus,” “Trigger EU & UK regulatory action,” and “Progress towards change in USA and support for STAR.”
That language matters because it strips away the polished rhetoric of public safety and reveals an organization openly pursuing pressure campaigns against named targets while simultaneously seeking the authority of a neutral nonprofit expert. A group whose recurring internal goal is to “Kill Musk’s Twitter” is not simply studying platform harms from a distance.
The memo says the quiet part out loud
The internal memo does not read like the notes of an independent research charity cautiously documenting online harms. It reads like the playbook of a coordinated influence operation measuring success in press hits, policy hits, legal leverage, donor cultivation, and regulatory outcomes.
One passage calls for reports on “metrics to do with press coverage and policy hits from recent reports.” Another describes a “big data point” in an antisemitism report “to drive profile growth.” Elsewhere, election-disinformation content is discussed as a way to “plant a flag,” while AI-related content is described as useful for “SEO - increasing traffic to CCDH” and “fundraising too.”
That is not ordinary mission language. It is campaign language.
Synthetic media for pressure politics
The sharpest contradiction emerges in CCDH’s AI planning. Publicly, CCDH presents itself as warning that AI systems are dangerous because they produce convincing synthetic content, lack guardrails, and can be weaponized against vulnerable people and democratic systems. Privately, the memo shows CCDH planning to do precisely that for strategic effect.
In the May 23, 2024 entry, CCDH lays out an “AI voice launch” titled Attack of the Voice Clones. The plan includes U.S. outreach to Senator Amy Klobuchar’s team for a quote or endorsement, U.K. outreach through the Cabinet Office, and EU/DSA outreach to the European Commission. It also includes specific discussion of how to use AI-cloned political voices with broadcasters, including the idea of a fake Sunak voice saying: “bad actors can get me to say anything, there are no safeguards.”
The memo states that CCDH had such recordings “ready to use for each of the eight main voices,” while also noting: “We can’t put out the dangerous ones.” It then adds: “We should make the point that the quotes are very convincing – and that is what makes the lack of safety and guardrails so dangerous.”
In plain English, CCDH was privately generating highly convincing AI political voice clones, deciding which ones were too dangerous to release, and using the rest to push lawmakers, regulators, and media outlets toward a preferred policy conclusion. This is not merely analysis of synthetic-media risk. It is active production and deployment of synthetic-media risk for advocacy purposes.
The jailbreak admission
Then comes one of the most revealing lines in the entire memo. In the same May 23 section, discussing “AI images for EU and UK politicians,” CCDH notes that the images had already been created for an NBC request and could be repurposed for broader media use. The memo continues: “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.”
That line should stop readers cold.
CCDH is publicly positioned as an organization warning that AI companies have failed to install adequate guardrails. Yet internally, it is explicitly discussing how it will jailbreak those tools, circumvent safety features, and then use the resulting outputs as “evidence” in a press campaign claiming the companies have not kept their promises on guardrails.
The memo even sketches the release strategy: an “EU/UK press only hit” on June 5, framed as a warning to voters that major AI platforms had not put in the promised protections. So the same organization criticizing the availability of dangerous outputs was deliberately provoking those outputs to generate a media-ready case for regulation.
Testing political smears with AI
The election-disinformation planning is equally revealing. On January 8, 2024, CCDH discusses how to “intervene in a year of big elections,” including AI’s role in disrupting elections and the press appeal of AI-related case studies. One particularly blunt line reads: “funders and journalists love it. 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.”
That is not passive monitoring. It is internal discussion of using AI systems to generate extreme political imagery involving named U.S. candidates in order to prove a point about what those systems allow.
Another participant explains the strategic advantage clearly: “next time there is a popular deep fake video, we can refer back to the fact we had pointed out that these deep fake sites were not ready and we had said so… this would be a good thing to have in our back pocket.” The point was not only to observe a problem, but to pre-position a narrative asset for future crises.
Child safety, list building, and lobbying lines
The memo also sheds light on how CCDH discussed parent-focused and child-safety content. Publicly, CCDH has built much of its authority on claims that it protects children from platform harms, self-harm content, and manipulative digital systems. Internally, the memo shows that a “Parents’ Guide” they were co-developing with Prince Harry’s Archewell Foundation was also valued as a conversion tool.
One line states that downloadable PDFs are “very useful for generating leads.” Another says the guide needs to be “behind an email sign up.” A further remark notes a “66% conversion rate at moment for the guide.” In the same discussion, the memo raises the need to check lobbying limits and, “if necessary register a c4 ASAP.”
None of this proves the child-safety concerns are insincere. What it does show is that material publicly framed as a moral response to harms facing children was also being assessed internally through the lenses of audience capture, conversion, and lobbying exposure.
This is where the child-safety framing collides with one of the most revealing episodes of the entire saga: the Kids Online Safety Act (KOSA). CCDH publicly endorsed KOSA, with CEO Imran Ahmed praising the bill for forcing platforms to be “safe by design” and touting its “duty of care” obligations to protect children. On its face, that is exactly the posture you would expect from a children’s-safety organization.
But the same internal documents that exposed CCDH’s lead-generation calculus also surfaced under a far more damning banner: the leaked “Kill Musk’s Twitter” strategy memo, in which CCDH discussed how to trigger regulatory action against X and degrade the platform’s viability. CCDH has been implicated in aggressive campaigns to deplatform dissenting voices, including Elon Musk himself — the very platform owner whose company would later line up behind a CCDH-backed bill.
And that is the extraordinary irony. As I documented in my earlier alert, “Elon Musk’s X Corp, under his leadership and that of CEO Linda Yaccarino, recently endorsed the Kids Online Safety Act (KOSA)” — the same legislative vehicle championed by an organization that had internally strategized to kill his platform. A self-described “free speech absolutist” and his CEO ended up endorsing a bill backed by the group most aggressively working to regulate, defund, and deplatform X. Whether through goodwill, political pressure, or simply not connecting the dots, Musk and Yaccarino risked empowering the very censorship architecture CCDH had been building — using the moral shield of child safety as the entry point.
The lesson of this section is not that protecting children is a pretext. It is that “child safety” can function simultaneously as a sincere cause, a lead-generation engine, and a regulatory wedge — and that even sophisticated operators like Musk and Yaccarino can be maneuvered into endorsing a framework that cuts directly against their own stated principles.
A campaign operation, not a neutral referee
The broader memo reinforces this pattern. There are references to lobbying in DC with “60 meetings on the Hill,” meetings with 16 congressional offices to discuss the Elon lawsuit and tech policy priorities, active monitoring of lobbying tracking, and briefing preparation for use “in Hollywood.” Celebrity networks such as CAA and UTA appear in the planning, along with discussions of endorsements, PSA circulation, and partner lists. Indeed, in 2024 CCDH hosted a DC event with government officials, NGOs, and even a British embassy official, the details of which are below.
At one point, CCDH even contemplates how to “attach our name” to Canada’s Online Safety Bill despite having “no actual staff in Canada.” At another, internal discussions emphasize how the “State of STAR” conference can showcase that multiple countries now have aligned legislation and that “showing a global movement will be powerful.”
This is the language of a transnational pressure network building legal, media, donor, and regulatory momentum in tandem. It is not the language of a detached umpire simply calling balls and strikes on digital harm.
Why this contradiction matters now
Florida’s lawsuit against OpenAI is likely to draw national attention because it merges AI safety, child harm, consumer deception, and public nuisance into one sweeping state action. If CCDH is going to be treated in that context as a trustworthy outside authority on AI ethics and guardrails, then its own internal approach to AI should matter too.
And what the memo shows is a profound contradiction. CCDH publicly condemns realistic synthetic media, weak AI safeguards, manipulative digital design, and dangerous outputs. Privately, it discusses creating convincing voice clones, generating political AI images, jailbreaking tools to bypass safety systems, and packaging those outputs for lawmakers, broadcasters, major newspapers, donor audiences, and election-cycle narrative advantage.
That does not look like a neutral guardian standing above the fray. It looks like a highly professionalized pressure operation, operating in effect as a foreign influence campaign, that wraps itself in the language of “safety” while using tactics that closely resemble the conduct it accuses others of. That, in turn, runs in parallel with Ahmed’s case in Ahmed v. Rubio, where he argues that U.S. officials are unconstitutionally punishing him—by revoking his visa—for the very advocacy and research that pressure governments and platforms to police speech. A stinging irony I have documented in detail below.
The real issue is not whether AI risks are real; they are. The issue is whether the organizations claiming the moral authority to police those risks are themselves operating honestly. In CCDH’s case, the gap between the public brand and the private memo is not small. It is the story.
If you want to learn more my own federal civil rights lawsuit against CCDH/Imran Ahmed, watch our recent live discussion below with our lawyer, Jeff Childers, who is also has the immensely popular Substack series: ☕️ Coffee & Covid News 🦠
Share the X thread dedicated to this article.
Endnotes
Complaint, Office of the Attorney General, State of Florida v. OpenAI Global, LLC et al., attached PDF: openai-filed-stamped-complaint.pdf. Used here for the complaint’s reliance on CCDH’s AI-harm research and its framing of ChatGPT as dangerous, deceptive, and unsafe for minors.
Portfolio Planning Monthly Meeting memo, attached PDF: portfolio-planning-monthly-meeting-notes-4.pdf. Used here for CCDH’s internal planning language, including “Kill Musk’s Twitter,” “Trigger EU & UK regulatory action,” AI voice-clone planning, AI-image jailbreaking, election-disinformation strategy, lead generation, and lobbying-related discussions.
We the People need to be ever vigilant against the evils of this Realm.
🙏🙏🙏🙏🙏
The evil never ends! It’s rather discouraging. On a lighter note, I love Jeff Childers and his substack. You have an excellent attorney and I wish you well in your lawsuit!!