Information control

Information control is one of those topics where almost every reader brings strong priors, and where those priors usually map onto partisan identification. This lesson tries hard to be even-handed because the structural mechanisms are real regardless of who is currently operating them. Today's victim is tomorrow's beneficiary; the institutional architecture outlasts any particular political alignment.

The base rate is this: powerful actors have always tried to shape information. The 20th-century governments of essentially every ideological stripe — democratic, authoritarian, communist, fascist, theocratic — invested heavily in information operations. What's new in the 2010s-2020s is who holds the choke points (platforms, not broadcasters) and the speed and scale at which influence can be applied.

The seven major mechanisms

1. Direct state censorship. The simplest case: government tells media what can't be published, jails violators. Modern examples include China's Great Firewall (extensive, automated, integrated with surveillance), Russia's wartime media laws (which have made independent journalism functionally illegal since 2022), Iran's internet shutdowns during protests, and a long list of dozens of less-noticed cases globally. The US and EU have weaker formal versions (FCC content rules, German hate speech laws, UK Online Safety Act), but the structural mechanism is the same: state authority defines the boundaries of permissible speech.

2. Platform moderation at scale. Meta, Google (YouTube), Twitter/X, TikTok, and Apple set the rules for what billions of people can publish and find. These are private companies operating under enormous political pressure from multiple governments and stakeholder groups. The decisions get made by trust-and-safety teams whose policies are nominally public but whose enforcement is opaque, inconsistent, and consequential. The Twitter Files (2022-23) and similar internal-document leaks at other platforms have documented this in detail, including specific cases of government communication with platforms about content decisions.

3. Algorithmic downranking ("shadow banning"). Not removal — just lower visibility. The most consequential form of platform moderation because it is essentially undetectable to users. A post that gets 1/10th its expected reach has effectively been censored, but neither the author nor their followers can tell. The platforms' own documented enforcement guidelines (Meta's "borderline content" policies, YouTube's "recommendation suppression") confirm this category exists; the specifics of how it's applied to any individual case are not transparent.

4. Deplatforming. Removing accounts entirely. The high-profile cases (Trump from Twitter/Facebook in 2021, Alex Jones from multiple platforms in 2018, various journalists and activists across ideological positions) show the mechanism. The systemic version — coordinated cross-platform action against the same set of actors at the same time — is more consequential. Coordinated enforcement has happened in both directions (against right-wing extremists post-Jan 6; against some left-wing journalists during Israel-Gaza coverage in 2023-24).

5. Payment and infrastructure squeezing. Stripe, PayPal, Visa, Mastercard, ACH banks, cloud hosts, domain registrars, app stores — each represents a chokepoint where service can be cut. The most extreme version: the 2022 Canadian trucker protest case, where the government invoked emergency powers to freeze the bank accounts of protest organizers and donors without judicial process. Less extreme but more common: payment processors that drop adult content, certain types of political speech, or specific industries (gun retailers, cannabis businesses). The First Amendment doesn't apply to private payment infrastructure; the practical effect is the same as censorship for those affected.

6. Narrative warfare and "flooding the zone." Active production of disinformation at industrial scale to drown out signal. Russian, Chinese, Iranian, and US state-affiliated operations have all been documented running troll farms, fake news sites, and bot networks. The generative AI revolution has made this dramatically cheaper. A 2025 disinformation campaign that produces convincing video, text, and voice at near-zero marginal cost is qualitatively different from anything that existed five years ago.

7. Legacy media consolidation. The pre-internet press model — independent local newspapers, broadcast networks with significant editorial independence — has been progressively replaced by consolidated holdings (Sinclair Broadcasting locally; Murdoch's News Corp; Carlos Slim's NYT stake; Bezos's Washington Post; private equity ownership of dozens of major dailies). Each major outlet still has reporters doing real work; the editorial direction is set by a small number of owners and a smaller number of editors who answer to those owners. This isn't censorship in any classical sense; it is structural narrowing of what gets covered and how.

The Section 230 / DSA situation

The legal substrate matters. In the US, Section 230 of the Communications Decency Act (1996) gives platforms immunity from liability for user-posted content. Without it, platforms would either have to moderate everything aggressively (eliminating most speech) or be sued into oblivion for any defamatory or harmful post. With it, they have enormous discretion to moderate or not as they choose, with little legal exposure either way.

In the EU, the Digital Services Act (2024 enforcement) imposes obligations on "very large online platforms" to address illegal content, transparency requirements, and risks to fundamental rights. The penalties for non-compliance are real (up to 6% of global revenue). The DSA represents one of the most ambitious attempts globally to regulate platform power; whether it achieves its goals without unintended consequences will be one of the most-watched policy experiments of the late 2020s.

In China, the model is essentially the opposite: platforms are explicitly required to censor and to cooperate with state surveillance, and they do. The middle ground — democratic accountability over platform power without state capture — has not yet been institutionally solved anywhere.

The generative AI inflection

Through 2022, producing convincing fake images, text, and video at scale required money and skill. After 2022 (Stable Diffusion, ChatGPT, DALL-E, then the wave that followed), it requires neither. A teenager with a laptop can produce influence campaigns that would have required state resources a few years earlier.

This cuts both ways: empowering legitimate voices that couldn't previously produce media; and empowering bad actors at unprecedented scale. The platforms have not figured out how to handle this. Government regulation is years behind. The information environment of the next decade will be shaped by how this is or isn't governed.

What this actually looks like, lived

For most people, information control isn't experienced as censorship. It's experienced as a slow narrowing of the visible world. The story you would have noticed gets buried three pages down in search; the voice you would have found gets recommended less; the protest you would have heard about isn't covered; the angle that would have changed your mind isn't algorithmically promoted to your attention.

This is why the "censorship vs. free speech" framing on both sides of the political aisle misses the point. The question isn't whether some specific post got removed (it usually didn't). The question is how the information environment that shapes what people believe has been systematically narrowed — by platforms, by states, by the economics of attention, by the consolidation of who owns the channels. That's the real story, and it doesn't have a simple ideological hero.