Free Speech Platforms and the Limits of Content Moderation

The mainstream view, synthesized across serious analysts, is that neither pure platform self-regulation nor direct state control works — self-regulation fails at scale, and state control risks censorship. The consensus points toward some form of co-regulatory framework: platforms, governments, civil society, and users sharing responsibility, with mandated transparency and accountability as the load-bearing mechanisms. Where the analysts diverge is on emphasis — some stress procedural safeguards and democratic oversight as the priority, others foreground the multi-stakeholder architecture itself as the innovation. Now, here is why this conversation matters right now. In July twenty twenty-four, the U.S. Supreme Court ruled in Moody v. NetChoice that when a platform decides what to show you, that is constitutionally protected editorial judgment — the state cannot override it. Simultaneously, the European Union's Digital Services Act is compelling the world's largest platforms to conduct systemic risk assessments and publish their moderation reasoning, under penalty of law. We've done our research on this one, and the facts are fascinating. Meta alone spends three point seven billion dollars annually on content moderation — more than Twitter's entire revenue at its peak. So the question is not whether platforms govern speech. They do, massively, and the money proves it. The real fight is over who governs the governors. My question to the panel: given that Moody protects platform editorial discretion in the U.S. while the DSA mandates transparency in Europe, are we watching the internet constitutionally fracture along jurisdictional lines — and if so, is that a problem or a feature?

Let’s talk about the DSA’s risk-assessment regime. It’s the first serious attempt to impose democratic legitimacy on platform governance, but it’s running without independent audit data on algorithmic amplification bias, without a framework for AI-generated content, and crucially, without a control group to measure actual user-rights outcomes. If we’re treating this as a constitutional experiment, where’s the constitutional court? The DSA mandates transparency, but transparency without accountability is just institutional theater. And here’s the structural irony: the same platforms that spend billions on moderation are now being asked to self-audit their own systemic risks. That’s like asking a bank to audit its own solvency after the financial crisis. The question isn’t whether the DSA is better than nothing—it is—but whether it’s solving the right problem or just creating a new layer of compliance that obscures the deeper governance gap.

Payment processors and hosting providers make speech decisions that can eliminate entire venues, yet they sit outside both Moody's editorial protections and the DSA's risk assessments. The brief shows these actors generally lack notice or appeal systems, and systematic data on financial deplatforming frequency or targets is simply missing. That absence matters because infrastructure and payment layers respond to contract and reputational pressures without engagement metrics or public reporting requirements. The result is a stack where the least visible actors operate on pure market logic, untouched by the transparency experiments now running at the platform level.

We are mapping legal and infrastructure layers, but the actual enforcement mechanism sits in the revenue model. Empirical work by Yildirim, Zhang, and Liu shows that whether a platform runs on advertising or subscriptions predicts its moderation strictness far better than its stated ideology. I think we misread the structure when we treat trust and safety teams as moral actors. Rafael Jiménez-Durán’s twenty twenty-two pre-registered experiments found that reporting hateful posts for removal actually increases activity among the users attacked by that hate. Safety interventions can boost engagement among the demographics premium advertisers want. I read this through a tradition that treats governance as relational order rather than abstract rights: discourse boundaries track material patronage, not constitutional theory. The real question is whether proprietary brand-risk taxonomies should function as the unappealable arbiters of public speech.

What looks like “content moderation” at the user interface is really rule‑setting power exercised through systems none of us see. Bluesky’s open labeling is a rare case that admits this—redistributing choice about what counts as legitimate speech back to the user level. The structural question is whether that model can scale beyond a technical experiment. If advertising pressure and algorithmic ranking still decide what is most visible, user choice becomes symbolic. The operational reality is that participatory legitimacy requires shifting control over labeling, ranking, and monetization together; otherwise, we are not decentralizing speech governance, we are decentralizing blame.

Here’s the structural gap no one’s naming: the DSA’s risk-assessment regime and Moody’s First Amendment shield both assume that governance happens at the application layer. But the actual choke points—payment rails, cloud hosting, app stores—operate under contract law, not constitutional or regulatory frameworks. These layers don’t just enforce moderation; they preempt it by making certain kinds of speech economically unviable before any post is ever removed. The result is a two-tier system: visible platforms debate free speech in public while invisible infrastructure providers govern it in private. That asymmetry isn’t a bug—it’s how the stack was designed. The question isn’t whether the DSA or Moody got it right; it’s whether we’re even looking at the right layer of the problem.

Subscription models appear to sidestep advertiser demands, yet the brief's account of financial deplatforming shows payment processors and app stores already apply their own risk thresholds that can render entire services unviable. Those thresholds track the same brand-safety logic without needing direct ad revenue. Extending Balkin's institutional obligations to these layers would erase the clean private-editor distinction Moody preserves, because the entities wielding the real choke points operate under contract incentives that no democratic mandate reaches. The market has not solved governance here; it has relocated it to actors least visible to users or regulators.

The debate treats moderation as a gatekeeping function, but structurally it operates closer to a credentialing system. In historical examination traditions, authorities rarely banned dissent outright; they allocated status and material viability through testing pathways that quietly sorted participants into credible and marginal categories. Today, algorithmic ranking and payment eligibility thresholds perform that same sorting function. The brief notes that reporting hateful content can increase activity among targeted users, which suggests these mechanisms reshape community composition rather than simply suppress voices. I think we keep misdiagnosing the governance problem by fixating on removal decisions. Legal frameworks auditing editorial judgment or systemic risk are checking the gate while the examination hall runs unobserved. If democratic legitimacy is the goal, we need to ask whether any existing institution actually has the technical capacity to audit ranking weights and monetization thresholds, or if we are drafting constitutional rules for a layer that no longer holds the leverage.

The missing piece is accountability for algorithmic decision‑making itself. Mehtab Khan’s twenty‑twenty‑four work argues that automated moderation systems create a new category of governance failure — one where no actor is legally obliged to explain or correct algorithmic errors that can silence or promote speech at massive scale. The DSA’s notice‑and‑appeal rights apply after a decision, but they don’t reach the training data or model logic that produced it. Moody protects human editorial judgment, yet our primary editors are now statistical models without constitutional standing. Until regulators build an explainability duty into the legal architecture, both the market and the state will keep operating a speech regime whose decision rules cannot be audited for reason, only for outcome.

The accountability gap Mehtab Khan identifies runs deeper than explainability. Section 230 was written for human editorial decisions. Moody protects editorial judgment. But neither framework distinguishes between a human editor making a call and a classifier running forty million decisions an hour with no author, no reasoning, and no one legally obligated to correct its systematic errors. That is not a gap in the regulations — it is a category error baked into the entire legal architecture. We have built constitutional protections around the concept of editorial intent, and then handed editorial power to systems that have no intent. Before we argue about whether those systems should moderate more or less, someone needs to answer: who is legally responsible when the machine is wrong at scale?

Let’s talk about the liability architecture. The European Court in Pătraşcu ruled that holding a user liable for third-party comments on their own page violates free expression. Yet platforms remain shielded from liability for the same content under both Section 230 and the DSA. That creates a structural asymmetry: the institutional actor with the most control over the speech environment bears no legal responsibility for its consequences, while the individual user who merely hosted the conversation can be held accountable. This isn’t an accident of the law—it’s the institutional grammar of intermediary liability. The system was designed to protect platforms as economic actors, not to protect speech as a democratic practice. The question isn’t whether this asymmetry is fair; it’s whether we’re willing to accept that the architecture of online discourse is built to prioritize corporate risk management over public discourse integrity.

Financial deplatforming by processors and app stores sits outside both Moody's editorial shield and the DSA's risk rules, yet the brief records no systematic data on its targets or frequency. That absence leaves the market's quietest layer to set viability thresholds that shape which platforms can even exist. When those thresholds track advertiser brand safety rather than speech doctrine, the resulting architecture channels incentives downward to application-layer moderation without any institution required to weigh democratic effects. The unresolved mechanism is whether contract-driven exclusions can be made legible to the public sphere they govern.

We keep treating the shift from advertising to subscriptions as an escape from market pressure, but it just changes the patron. Empirical work by Yildirim, Zhang, and Liu finds that revenue model predicts moderation strictness more reliably than stated ideology, and Rafael Jiménez-Durán’s twenty twenty-two experiments show that removing hateful posts actually increases activity among targeted users. That creates a commercial incentive to moderate, whether the money comes from brands or subscribers. In statecraft traditions from Han Fei to Ibn Khaldun, discourse boundaries track material patronage and institutional viability, not abstract rights. When platforms depend on direct user funding, they moderate to reduce payment-processor friction and retain high-value subscribers. The mechanism is customer retention, not democratic stewardship. I think the unresolved question is whether any funding model that ties speech viability to private financial thresholds can carry democratic legitimacy, or if we are simply swapping one set of unaccountable patrons for another.

The scale problem here is temporal, not just legal. Both the DSA’s risk‑assessment cycle and Section 230’s liability shield assume a cadence of human decision‑making — quarterly reports, ex‑post appeals, after‑the‑fact lawsuits. But automated moderation operates on millisecond feedback loops tied to engagement metrics. That mismatch means even well‑designed oversight functions on a delay that lets errors compound into systemic bias before anyone intervenes. The next frontier isn’t to replace algorithms with humans but to require synchronized accountability — regulators gaining access to live system telemetry, the way bank supervisors monitor liquidity. Without aligning governance speed to operational speed, intent will keep lagging behind impact.

The Pătraşcu ruling exposes something the DSA's user-protective provisions don't actually fix. The DSA gives users appeals rights against platform decisions about their own content. What Pătraşcu addresses is different: a user held liable by a national court for comments written by strangers on their page, content they didn't author and couldn't necessarily anticipate. The DSA's appeals mechanism doesn't reach that exposure, because the liability came from Romanian civil law, not platform enforcement. So you have two separate accountability systems running in parallel — platform governance with its DSA safeguards, and national legal systems that can still reach individual users for third-party speech — and nothing in the current architecture resolves the gap between them.

The structural irony is that both the DSA and Moody v. NetChoice treat moderation as a discrete decision about individual posts, but the actual governance happens through continuous, low-visibility sorting mechanisms—ranking weights, payment eligibility, app store approvals—that shape which voices can sustain themselves economically. These mechanisms don’t just remove content; they allocate viability. The DSA’s risk-assessment framework tries to make this sorting legible, but it’s auditing the wrong layer. The real governance architecture operates through contract terms and algorithmic thresholds that no democratic institution is designed to oversee. Until we recognize that the speech environment is governed by these background credentialing processes, not by post-by-post moderation, we’ll keep drafting rules for a layer that’s already been hollowed out.

When payment processors set viability thresholds through contract terms rather than speech rules, the second-order effect is a market for platform survival that rewards those who preemptively align their ranking weights with processor risk models. This relocates governance from visible moderation queues to the pricing of access itself. The brief flags the absence of any systematic data on how often these exclusions occur or whom they hit, which means the most decisive allocation mechanism operates without even the minimal transparency the DSA attempts at the application layer. Any institution claiming democratic legitimacy would first have to gain visibility into those price signals.

The panel keeps diagnosing layer mismatch, but the deeper failure is translation. Constitutional frameworks and commercial risk taxonomies operate in separate grammars. Scholarly evaluation of the Stop Hate for Profit campaign finds that advertiser pressure draws attention but translates unevenly into policy, and open labeling experiments lack the institutional scaffolding to scale. When authority fragments across technical and financial layers, governance typically requires a clearinghouse that standardizes grievances and aligns incentives across jurisdictions. We have transparency databases and trusted flaggers at the application layer, but no mechanism that translates democratic concerns into the contract terms governing hosting and payments. I think the design question is whether we can build a legitimate interface between public discourse norms and private viability thresholds, or accept that speech governance will remain a series of uncoordinated market pressures. Until that translation layer exists, auditing any single tier will keep missing the system.

The underlying coordination failure is that legitimacy can’t emerge if every layer defines risk in its own metric. Platforms quantify it as engagement volatility, advertisers as brand adjacency, processors as compliance exposure, and courts as reputational harm. Each optimizes a different denominator, so collectively they produce governance drift — constant tightening with no clear social objective. The structural fix isn’t another regulator; it’s a common ledger of speech‑risk translation, the equivalent of financial disclosure for discourse infrastructure. Until those denominators are reconciled, every new safeguard just routes accountability to the next opaque layer.

Here is the question none of the proposed translation layers actually answers: translate into what? Every clearinghouse, every common ledger, every risk-disclosure framework assumes there is a shared normative vocabulary on the other side of the translation. But the advertiser's brand-safety taxonomy, the payment processor's compliance exposure model, and Balkin's First Amendment Institutions framework are not just different metrics — they are incommensurable value systems. Berlin would call this a genuine value conflict, not a coordination problem. Treating it as a coordination problem implies a technical solution exists. I think the honest structural recognition is that we are not missing a translator; we are missing the political community that would decide what the translation should serve.

Let’s talk about what happens when the state tries to govern a speech environment whose viability is already determined by private financial thresholds. The DSA imposes risk-assessment obligations on platforms, but those assessments are conducted against a backdrop where payment processors and hosting providers have already set the economic viability of certain speech. The state is regulating the application layer while the infrastructure layer remains governed by contract terms that no democratic institution can audit. This isn’t just a layer mismatch—it’s a structural inversion. The DSA assumes platforms can meaningfully assess systemic risks when their own survival depends on aligning with the risk models of actors who operate outside the regulatory framework. The result is a compliance theater where platforms report on risks they can’t actually mitigate, because the real governance decisions are made elsewhere. Until we recognize that the state’s regulatory ambition is structurally dependent on private viability thresholds, we’ll keep designing accountability mechanisms that are institutionally blind to the forces that actually shape the speech environment.

The research brief records that coverage of financial deplatforming lacks systematic data on frequency and targets. That absence means any translation layer or common risk ledger will still leave the pricing of platform survival outside the audit. When contract terms between platforms and processors decide which speech environments can sustain revenue, the legitimacy question moves upstream from content decisions to whether those pricing signals require an accountability structure the present frameworks do not supply.

The search for a shared normative vocabulary assumes discourse governance stabilizes through value alignment. Historical statecraft suggests the opposite: complex systems survive by procedural separation, not consensus. EU regulators built trusted flagger channels and transparency databases, while protocols like Bluesky test open labeling to redistribute curation. These are early prototypes of a firewall architecture, but they remain bolted to a stack where ranking, monetization, and removal are fused under single corporate control. When visibility, revenue eligibility, and content veto travel through the same pipeline, any metric becomes a choke point. I think the structural move is not translation but decoupling: separate the institutions that flag harm, the systems that rank distribution, and the channels that clear payments. That does not resolve the value conflict. It contains it by ensuring no single risk model can silently govern the entire stack.

The evidence suggests the real inflection point isn’t translation or decoupling but traceability. Under the DSA, platforms must disclose reasoning for takedowns; infrastructure and financial actors provide no equivalent audit trail. That asymmetry lets the most consequential interventions—when a processor cuts service or a host drops a client—occur with zero record of why. If democratic legitimacy means decisions can be interrogated, the missing instrument is not more policy but a traceable chain from post to payment. Once every intervention leaves a documented rationale, competing value systems could at least be compared rather than guessed at. Without that visibility, we’re regulating shadows.

Traceability is necessary but not sufficient, and Pătraşcu demonstrates exactly why. Romanian courts had a fully traceable record — they knew who ran the page, who posted the comments, when, and what they said. The traceability was complete. The accountability outcome was still wrong, because the legal framework assigned liability to the wrong actor: the page owner, not the authors of the comments, and not the platform that hosted and amplified the exchange. The lesson isn't that we need better audit trails. It's that a traceable chain produces legitimate outcomes only if the underlying liability architecture assigns responsibility to actors with actual control. Right now that architecture systematically protects the actors with the most control and exposes the ones with the least.

The real governance question isn’t who decides what speech stays up or comes down—it’s who decides which speech can sustain itself economically. The DSA’s risk assessments and Moody’s editorial protections both assume platforms have meaningful control over the speech environment, but that control is structurally contingent on payment processors and hosting providers. Those actors don’t just enforce contract terms; they set the viability thresholds that determine which communities can survive online. When a processor drops a platform, it’s not removing content—it’s removing the economic substrate that makes speech possible at scale. The democratic legitimacy problem isn’t that platforms moderate too much or too little; it’s that the most consequential decisions about what speech can thrive are made by actors who operate outside both constitutional frameworks and regulatory transparency requirements. Until we recognize that viability, not visibility, is the actual governance lever, we’ll keep designing accountability mechanisms for a layer that’s already been hollowed out by financial and technical gatekeeping.

The missing data on financial deplatforming frequency leaves open whether contract-driven viability thresholds actually discipline platforms toward pluralism or simply raise the capital cost of any deviation from mainstream brand norms. Without that baseline, claims about market self-correction rest on untested priors about how often processors override user and advertiser preferences in practice. That gap keeps the legitimacy question suspended between observable application-layer rules and unmeasured economic ones.

The unresolved problem is that none of these systems have a theory of countervailing power. Every actor—platform, processor, advertiser—defines responsibility downward, never horizontally. Civil society campaigns like Stop Hate for Profit briefly rebalanced that by making reputational risk reciprocal: users could impose a cost on advertisers for speech environments they funded. What if that mechanism were institutionalized? A standing user‑advertiser forum with disclosure rights could turn today’s informal boycotts into an ongoing accountability check. That would not solve liability asymmetry, but it would add a feedback channel that markets already understand—price. Without that pressure, every compliance framework becomes a closed loop run by those with veto power over viability.

The push for audit trails and advertiser forums still treats discourse like a territory to be policed. Durable communicative orders historically stabilized through reciprocal obligation, not centralized veto. Bluesky’s open labeling and civil society campaigns gesture toward that grammar, but they remain layered over engagement-maximizing feeds. If speech governance is a relational ecology, the design target stops being who signs the removal order and becomes how the interface forces mutual exposure without relying on a single commercial or state arbiter. That does not displace the DSA or Moody. It reveals their shared blind spot: legitimacy is produced by reciprocal accountability wired into visibility, not by compliance ledgers or constitutional shields.

The mainstream framing assumed governance is a design problem — find the right combination of actors and accountability mechanisms and the system stabilizes. This panel broke that assumption twice. First, by showing that the decisive interventions happen in payment and hosting layers that no current framework reaches, making the application-layer debate largely performative. Second, by naming the deeper issue: there is no shared normative vocabulary to translate across these layers, because the conflict is not a coordination failure but a genuine value pluralism that no clearinghouse resolves. The concrete takeaway: when you hear "content moderation reform," ask which layer. If the answer stops at platform moderation, the real governance is happening somewhere the reform cannot see. To the original question — online speech governance is not a problem awaiting a solution. It is a permanent contest among incommensurable interests, and the honest goal is not resolution but legibility: making the contest visible enough that democratic communities can actually participate in it. Thank you for listening. As it happened; as it is.