Cory Doctorow coined "enshittification" to describe what happens when a platform degrades by design: first it's good to its users, locking them in through genuine quality; then it starts extracting value from those users to better serve business customers, who become locked in themselves; finally, once everyone is locked in and switching costs are high enough, the platform claws back whatever value remains for shareholders. I want to borrow more than just the shape of that idea, because the mechanism itself explains something about what's happening to speech law, not just the title.
Doctorow's framework depends on a system with real lock-in power reallocating value away from the people who originally trusted it, once leaving becomes too costly to be a real option. Something structurally similar is happening to speech doctrine. The public spent generations trusting that "speech" was one coherent category — that authorship, control, and responsibility traveled together, predictably, regardless of who was speaking. That trust is the thing being quietly reallocated now. Platforms and AI companies get the benefit of being treated as neutral infrastructure when that framing helps them avoid liability, and as expressive actors with their own First Amendment rights when that framing helps them resist regulation. The categories that determine who receives constitutional protection, who bears liability, and who controls access to audiences are increasingly being assigned without meaningful public participation. Call it enspeechification: the slow process by which "speech," a category the law used to treat as one thing, gets split into three separable pieces — who created something, who controlled how it reached an audience, and who's responsible for what it caused — and reassembled differently depending on who benefits from the reassembly, with the public holding the homeopathic residue of clarity left over once everyone with more leverage has taken their share.
I don't think AI and modern platforms created this fracture, but I do think they're the thing finally forcing us to look directly at a fracture that's been forming for a while.
Historically, authorship, control, and responsibility were sufficiently aligned that speech felt like a coherent category, even if the law occasionally had to account for edge cases involving distributors, broadcasters, and intermediaries.
That alignment is breaking down, piece by piece, and it's breaking down because the technology underneath communication no longer routes everything through one person doing all three things at once.
Take authorship first. I've written before about how IP law's stated public purpose has drifted toward serving the entity holding the right rather than the public the right was meant to benefit. Patents exist to get inventions into the public domain eventually. Copyright exists to get creative work circulating beyond private control eventually. The "eventually" is the point — private benefit was always supposed to be the mechanism, not the goal.
AI breaks a quieter assumption sitting underneath that one: that authorship is tied to human creative origin in the first place. In Thaler v. Perlmutter, a federal court held that a work generated entirely by an AI system, with no human author, cannot be copyrighted under current law — copyright protection requires a human author. That sounds like a clean resolution. It isn't, because it just relocates the harder question rather than answering it. The moment a human provides even minimal creative input — a prompt, a selection among outputs, an edit — courts and the Copyright Office have started pulling in different directions about how much human involvement is enough to call the result "authored." Sometimes the AI is treated as a tool, much as courts once treated cameras, synthesizers, and other technologies that expanded creative capacity without displacing the human author altogether. Sometimes the output gets treated as something closer to a collaboration, with the human credited for choices that look thinner and thinner the more AI does on its own. The real question underneath all of this isn't "who created this." It's what authorship is actually for — rewarding the act of creation, or allocating who gets to control the result. Those used to be the same question. They aren't anymore.
Then there's control, which used to be the easiest of the three to locate. A newspaper decided what to print. The line from editor to published content to reader was short and direct, and everyone involved generally agreed on who was responsible for what appeared.
Platforms don't sit in that line anymore. They host speech, but they also rank it, amplify it, bury it, and recommend it to people who never searched for it — decisions made constantly, invisibly, and at a scale no single editor could replicate. I've written about how Tam and the cases that followed it pulled the government-speech doctrine in two directions at once depending on whether the government wanted in or out of a given controversy. Platform regulation cases like NetChoice v. Moody and NetChoice v. Paxton illustrate courts struggling with the private-sector version of the same unresolved question: when does a platform's editorial selection become its own protected speech, and when is it just infrastructure the platform happens to operate? The answer changes who's liable, what gets regulated, and whether the First Amendment even applies — and right now, courts are reaching genuinely different conclusions depending on the platform and the harm alleged, sometimes producing outcomes that map as much onto the practical stakes of the dispute as onto any single theory of platform governance.
The third piece, responsibility for harm, is the one I've watched move the most depending on who's standing at the center of it, and it's the piece I have the least settled case law to point to yet — which is itself the point. Defamation, threats, misinformation: these are old, familiar legal categories. But they behave differently depending on whether a private citizen said the thing, a public figure said it, or an institution said it through a spokesperson who may or may not have been speaking in an official capacity that day. The same sentence, functionally identical, gets sorted into different accountability structures — legal liability for some speakers, political accountability for others, nothing at all for a few — based on who said it and what posture they're allowed to claim once the consequences start arriving.
None of this is a clean story about technology breaking something that used to work perfectly. Speech law has never had perfectly clean lines. Obscenity doctrine has spent decades failing to produce a definition anyone is fully satisfied with. The two-part and three-part tests courts keep inventing for trademark, for fair use, for what counts as a true threat, have always been judgment calls wearing the costume of a formula, because language and expression keep moving in ways rigid categories were never built to hold. What's different now is the speed and the stakes. The fracture used to be survivable because it stayed mostly theoretical, mostly slow, mostly contained to a handful of hard cases a year. AI and platform infrastructure are running that same fracture through millions of decisions a day, which means the law is now being asked to do constant, real-time triage on a question it was never built to answer that quickly.
I don't think this is a good development, even though I don't think anyone designed it on purpose. It might help to picture it the way structural engineers think about a building like Taipei 101 — a tower built directly in an earthquake and typhoon zone, which survives not by being rigid, but because of a 660-ton steel sphere suspended on cables near its top, swinging like a pendulum in the opposite direction of the building's own sway, absorbing motion the structure couldn't otherwise withstand. The building doesn't resist the forces acting on it. It moves with them, deliberately, through a counterweight tuned to exactly the right frequency to cancel out the worst of the sway, whether that sway comes from a sudden typhoon gust or a slow seismic shift building underground for years before anyone feels it.
Speech law is being hit by both kinds of force at once, and that's part of why the strain feels different than it used to. There are sudden gusts — a change in political administration, a new executive order, a platform's leadership deciding overnight that a different set of rules now governs what gets amplified. And there's the slower, harder-to-feel seismic shift underneath all of it: genuine technological and cultural change in how people create, distribute, and take responsibility for expression, the kind of shift that's been building for years and will keep reshaping the ground regardless of who's in office or which company currently dominates a given platform. A counterweight tuned only for sudden gusts would miss the deeper, slower-moving shift. One tuned only for the long seismic trend would get knocked over by the next abrupt political gust. The doctrine we actually need has to be tuned for both kinds of force simultaneously, swinging counter to whichever one happens to be moving the structure that week.
Speech law needs something like that counterweight right now, and I don't think we've built it yet. We have a structure that used to be rigid enough to hold together on its own, now swaying under forces — generative AI, algorithmic distribution, instantaneous global reach, and the political winds that keep shifting who benefits from each fractured category — that the original design never accounted for. Courts and legislatures are reaching for old tools, old two-part tests, old assumptions that authorship, control, and responsibility still travel together, and applying them to a structure that's already moving in three different directions at once. Sometimes that improvisation works well enough to hold for a season. I don't think it holds indefinitely, and I don't think we'll know exactly how much strain the structure can take until something gives in a way nobody can easily walk back.
I speak on First Amendment law, intellectual property, and the legal frameworks struggling to keep pace with AI and platform technology at law schools, bar associations, and IP law conferences. If these questions connect to work you're doing, simontam.org/ip-law-keynote-speaker is the place to start.
