Research: The Censorship-Industrial Complex — The Organizations, the Money, the Courts, and the Case for the Defense

Companion pages: The Twitter Files · The Trust & Safety Pipelines · Content Moderation · Lawfare and the lawfare tracker. The vehicle side of the same argument is Vehicle Control Infrastructure.

“Censorship-industrial complex” is a polemical name for a real structure: government agencies, university labs, think tanks, NGOs and rating firms that between 2018 and 2024 studied, scored and flagged online speech, often with public money, and passed the results to the platforms that decided what stayed up. Matt Taibbi and Michael Shellenberger put the term into circulation in their testimony to the House Weaponization Subcommittee on March 9, 2023. Taibbi told the committee that “for every government agency scanning Twitter, there were perhaps 20 quasi-private entities doing the same, including Stanford’s Election Integrity Project, Newsguard, the Global Disinformation Index, and others, many taxpayer-funded” (Taibbi testimony; Shellenberger testimony).

What follows is the record of who did what, the money that paid for it, what the courts decided, and the strongest case its defenders make.

The organizations

Stanford Internet Observatory and its two partnerships

The Stanford Internet Observatory was the hub. Its Election Integrity Partnership was created in summer 2020, according to the House Judiciary staff report, “at the request” of the Cybersecurity and Infrastructure Security Agency; the report quotes the DFRLab’s Graham Brookie saying in July 2020 that the EIP was “set up . . . at the request of DHS/CISA.” The EIP ran a Jira ticket system that took reports of election content, analysed them and passed recommendations to platforms; the staff report reproduces tickets such as “We recommend removing or labeling these tweets” (House Judiciary staff report, Nov 7 2023).

The Virality Project carried the model to COVID-19 in 2021. Its March 17, 2021 email to Twitter listed the categories of content it was tagging and asked “which of the following types of content you would like to be tagged on going forward.” One category read: “True content which might promote vaccine hesitancy. Viral posts of individuals expressing vaccine hesitancy, or stories of true vaccine side effects. This content is not clearly mis or disinformation, but it may be malinformation (exaggerated or misleading).” Its examples included a central New York school district closing after staff fell sick with vaccine side effects. The email marks the categories Twitter was “currently receiving” in bold; in the screenshot as DiResta published it, only “Standard vaccine misinformation on your platform” is bold. Twitter replied that, beyond the default escalations, the categories that stood out were “known repeat offenders” and “notable content that is not on the Twitter platform” (full email and reply, published by DiResta).

Taibbi published the email in Twitter Files #19 as Stanford’s Virality Project “recommending against ‘stories of true vaccine side effects’ and ’true posts that could fuel hesitancy’” (Taibbi on X; Reason). DiResta, who co-led the project, says Taibbi and Andrew Lowenthal “cut it in half”: the email asked which categories Twitter wanted, and Twitter did not pick that one (DiResta). Both readings rest on the same document. What it settles is that the project tracked true posts as a category, filed them under “malinformation,” and offered to tag them; what it does not show is Twitter asking for that category or acting on it at the project’s request.

The Observatory came apart in 2024. Alex Stamos had left in November 2023; Renée DiResta’s contract was not renewed in June 2024 (Platformer; Washington Post). Stanford’s position is that the projects “provided public factual findings” and had no control over moderation decisions (Stanford).

The Global Disinformation Index

The Global Disinformation Index, a London not-for-profit, rated news sites for “disinformation risk” and supplied exclusion lists that ad exchanges such as Microsoft’s Xandr used to keep advertising off them. Its December 2022 list of the ten “riskiest” US outlets included Reason, the New York Post and RealClearPolitics (Reason); Senator Grassley noted that every outlet on it was “conservative or right-leaning.” About $330,000 of US federal money reached it, through the National Endowment for Democracy and the State Department’s Global Engagement Center, which led Senator Grassley to write to Secretary Blinken on April 12, 2023 that “taxpayer money shouldn’t be used to assist foreign entities in suppressing speech in the United States based on viewpoint or religion” (Grassley letter). NED dropped it in February 2023 (Washington Times).

NewsGuard

NewsGuard sells credibility ratings of news sites. The Air Force awarded it contract FA864921P1569 for $749,387 starting September 7, 2021 (USASpending), after it won a Pentagon and State Department contest for detecting COVID-19 misinformation (NewsGuard). The House Oversight chairman demanded it disclose all its government contracts (House Oversight). See the NewsGuard profile.

The DFRLab and Hamilton 68

The Atlantic Council’s Digital Forensic Research Lab was the EIP’s other main partner. Almost $7 million of the Atlantic Council’s 2022 spending went to it, per its annual financial report (Brownstone Institute).

Hamilton 68, the Alliance for Securing Democracy’s dashboard of accounts it said were “linked to Russian influence activities,” was cited in hundreds of news stories. Twitter’s own review, published in the Twitter Files on January 27, 2023, found that apart from a few RT accounts the list was “mostly full of ordinary Americans, Canadians, and British”; Yoel Roth wrote internally that it “falsely accuses a bunch of legitimate right-leaning accounts of being Russian bots” (Taibbi on X; Reason). Twitter did not say so publicly. The Alliance’s account of the dashboard is its own fact sheet (Alliance for Securing Democracy).

The Aspen Institute’s Commission on Information Disorder

Co-chaired by Katie Couric, Chris Krebs and Rashad Robinson, the commission issued 15 recommendations in 2021 (Aspen Institute). Taibbi: “The Woodstock of the Censorship-Industrial Complex came when the Aspen Institute … held a star-studded confab in Aspen in August 2021 to release its final report” (Twitter Files substack).

CCDH and Graphika

The Center for Countering Digital Hate took $1.1 million of its $1.47 million 2021 budget from the Schwab Charitable Fund, a donor-advised fund that does not disclose the donor behind a grant; its 2024 revenue was $4.25 million and it reported no government funding (InfluenceWatch). Graphika, per Foundation for Freedom Online, has received nearly $7 million in Defense Department grants and contracts, and its April 2020 report dates its collection of COVID “conspiracy theories” to December 16, 2019 (Foundation for Freedom Online).

The money

• NSF. Foundation for Freedom Online (Mike Benz’s organization) counted $38.8 million in NSF “misinformation” grants and contracts in the first two years of the Biden administration: 64 grants totaling $31.8 million to 42 colleges and universities, plus two contracts totaling $7 million (Foundation for Freedom Online). The House Judiciary Weaponization Subcommittee’s interim staff report of February 6, 2024 documented NSF grants for AI tools to counter misinformation “at scale” (House Judiciary). One recipient, Meedan, holds an NSF award of $749,974 for “FACT CHAMPION,” a fact-checker collaboration tool (NSF awards API).
• The Global Engagement Center. About 120 staff and a $61 million annual budget; it ended by operation of law on December 23, 2024 after Congress stripped its extension from the defense authorization (CyberScoop). See the GEC profile.
• CISA. Its mis-, dis- and malinformation team worked directly with platforms; the House Judiciary staff report on CISA and The Intercept’s October 2022 reporting on leaked DHS documents are the main records (House Judiciary CISA report; The Intercept). See the CISA profile.
• “Malinformation.” CISA defined it as information “based on fact, but used out of context to mislead, harm, or manipulate”: a government threat category that includes true statements (CISA; DHS advisory council subcommittee report).

The courts and the Congress

Murthy v. Missouri (June 26, 2024). The Court ruled 6-3 that the plaintiffs lacked standing, without reaching the merits. Justice Barrett wrote for the majority; Justice Alito, joined by Thomas and Gorsuch, called it “one of the most important free speech cases to reach this Court in years” and wrote that “a coterie of officials at the highest levels of the Federal Government continuously harried and implicitly threatened Facebook with potentially crippling consequences” (opinion).

Hines v. Stamos. America First Legal’s suit against the EIP and Virality Project survived the first motions to dismiss in December 2024 and went to jurisdictional discovery (Volokh Conspiracy). On July 1, 2026 Chief Judge Terry A. Doughty, the judge who issued the 2023 Missouri v. Biden injunction, dismissed it without prejudice for lack of jurisdiction, agreeing with the defendants that “discovery has confirmed the absence of any factual basis to conclude the EIP or VP, let alone any particular Defendant, coerced any social media platform to censor Plaintiffs’ posts or accounts” (Memorandum Ruling, Doc. 222; Judgment, Doc. 223). The plaintiffs appealed (5th Cir. No. 26-30471, pending); the case is on the lawfare tracker.

The Disinformation Governance Board. DHS announced it in April 2022 and paused it in May; Secretary Mayorkas “terminated the Disinformation Governance Board and rescinded its charter effective today, August 24, 2022” (DHS).

The Weaponization Subcommittee released a 17,000-page final staff report on December 20, 2024 (House Judiciary).

2024-2026: the structure comes down

• June 2024: SIO’s research staff dispersed; DiResta’s contract not renewed.
• December 23, 2024: the GEC ends.
• April 18, 2025: NSF states it “will not support research with the goal of combating ‘misinformation,’ ‘disinformation,’ and ‘malinformation’ that could be used to infringe on the constitutionally protected speech rights of American citizens”; some $328 million in grants had been cancelled that month (NSF; Nextgov).
• December 23, 2025: the State Department imposes visa restrictions on five people Secretary Rubio called “leading figures of the global censorship-industrial complex”: former EU commissioner Thierry Breton, CCDH’s Imran Ahmed, GDI’s Clare Melford, and HateAid’s Anna-Lena von Hodenberg and Josephine Ballon (Just the News; The Telegraph on X).
• Ahmed’s case. Ahmed, a US green-card holder, sued; Judge Vernon S. Broderick barred his arrest or detention on December 25, 2025 (JURIST). On September 8, 2026 Judge Loretta Preska dismissed the suit, holding that an appellate ruling left her without jurisdiction, and kept the restraining order in place for five business days to allow an appeal (Reuters via Internazionale).

Europe

The EU’s Digital Services Act is the most comprehensive content regime in force. The Commission’s two-year report counts 165 million moderation decisions appealed through platforms’ internal systems, almost 50 million of them reversed (European Commission). The Commission fined X EUR 120 million under the DSA in December 2025, and in February 2026 X filed what ADF International, which backs it, calls the first legal challenge to a DSA fine, at the EU General Court (European Commission; ADF International); the defense of the fine is that its three charges concern verification and transparency, not speech (Tech Policy Press).

The argument: a loop with no off switch

The Ratchet’s argument from this record is structural, not a charge of bad faith. Every participant was paid to find the problem: grants funded research that found the threat urgent, the findings justified the programs, the programs funded more research. The Hamilton 68 case is the clearest instance, a dashboard that labelled ordinary accounts as Russian for years while the platform that knew kept quiet. And the structure did not correct itself from inside: what ended it was discovery, congressional subpoenas, the Twitter Files and a change of administration.

The case for the defense

The strongest version of the other side, in its own terms:

• The foreign threat is real. In September 2024 the Justice Department indicted three IRGC officers for hacking the Trump campaign and trying to pass the stolen material to the Biden campaign and the press (DOJ); the Senate Intelligence Committee’s bipartisan report documents the Internet Research Agency’s 2016 campaign (Senate Intelligence Committee).
• The researchers say they did not censor. The UW Center for an Informed Public, co-founded by Kate Starbird: “CISA did not found, fund, or otherwise control the EIP. CISA did not send content to the EIP to analyze, and the EIP did not flag content to social media platforms on behalf of CISA”; its CISA collaboration ran from July to November 2020, under a Trump appointee, and “was reviewed and approved by Trump Administration attorneys” (UW CIP).
• A court looked. After discovery, the Hines ruling above found no factual basis for coercion by the EIP or VP; Renée DiResta’s account of the case is on her site (DiResta).
• The measured effects cut both ways. Exposure to the IRA’s 2016 Twitter campaign was concentrated (1% of users accounted for 70% of exposures) and the authors found “no evidence of a meaningful relationship between exposure to the Russian foreign influence campaign and changes in attitudes, polarization, or voting behavior” (Eady et al., Nature Communications, 2023). From inside the misinformation field, Altay, Berriche and Acerbi write that “alarmist narratives about online misinformation continue to gain traction despite evidence that its prevalence and impact are overstated” (Social Media + Society, 2023), and Budak, Nyhan, Rothschild, Thorson and Watts find “low exposure to false and inflammatory content that is concentrated among a narrow fringe” (Nature, 2024). Both sides cite these.
• Moderation works on its own terms. Reddit’s 2015 bans cut hate-speech use by the affected users by more than 80% (Chandrasekharan et al., 2017); removing hate organizations changed their audiences’ behavior (Thomas and Wahedi, PNAS, 2023).
• The scholarly defense treats moderation as systems governance rather than case-by-case censorship (Douek, “Content Moderation as Systems Thinking”), and Techdirt’s Mike Masnick argued the leaked DHS documents did not show censorship (Techdirt).

What would settle the open question, whether the arrangement amounted to government coercion, is a merits ruling. Murthy and Hines both ended on standing and jurisdiction, so no court has given one.

Related

• Profiles: Stanford Internet Observatory · Election Integrity Partnership · Renée DiResta · Alex Stamos · Yoel Roth · Global Disinformation Index · NewsGuard · DFRLab · CCDH · Graphika · Global Engagement Center · CISA · Virality Project · Kate Starbird
• The critics: Matt Taibbi · Michael Shellenberger · Mike Benz · Andrew Lowenthal
• Research: The Twitter Files · Trust & Safety Pipelines · Watching the Watchers · Lawfare · Content Moderation · Deplatforming
• Elsewhere: the security-conference and open-source governance side of the same pattern is on troll.fan (conferences; governance).