For nearly two decades, social media grew with almost no regulatory friction. Only now, after well-documented harm to mental health, political discourse, and children’s wellbeing, governments are catching up, passing age-verification laws, algorithm-transparency rules, and youth-protection statutes.
The pattern is familiar: underestimate the impact, let the technology embed itself into daily life, and only legislate once the damage is diffuse, entrenched, and expensive to undo.
The question worth asking is whether we’re doing the same thing with LLMs. Increasingly, researchers, policymakers, and even former social media executives think we are. The argument has moved from op-ed speculation into peer-reviewed scholarship over the last two years.
The core parallel
Nathan Sanders & Bruce Schneier, “Let’s not make the same mistakes with AI that we made with social media” (Harvard Kennedy School Belfer/Berkman Klein Center, 2024) is the most-cited articulation of this argument. Schneier — a well-known security and policy researcher, not a commentator — identifies five structural similarities driving both industries: advertising, surveillance, virality, lock-in, and monopolization. The argument isn’t just that the harms rhyme, but that the business model producing them is the same one, which is why the same regulatory tools (transparency requirements, antitrust, liability) apply to both.
Napoli & Adi, “On moving fast and breaking things.. again: social media’s lessons for generative AI governance” (Information, Communication & Society, 2025) makes the more academic version of this case. The paper’s target isn’t just specific harms but the governance failures that let those harms compound: the treatment of “innovation” as a governing principle in itself, the limits of self-regulation as a substitute for real oversight, and the absence of any dedicated regulatory body for either technology. Their point — stakes may be higher this time, not lower — is worth sitting with.
“Governing AI Companionship: Lessons from Social Media Regulation” (ACM FAccT 2026) narrows the parallel to the sharpest possible case: AI companion apps. It connects a wave of documented harms — including suicides linked to AI companions like ChatGPT and Character.AI, and evidence of addictive usage patterns — to the exact regulatory playbook built for social media (notice-and-takedown, age verification), while noting enforcement is already struggling against the pace of the technology. This is close to a direct empirical test of the hypothesis: same harm category (youth mental health, addictive design), same proposed fix, applied one cycle later.
An earlier, more policy-oriented paper — “Generative AI Regulation Can Learn From Social Media Regulation” (arXiv, 2024) — makes a narrower, more practical version of the same case, focused specifically on content moderation and bias, and explicitly frames the goal as saving “effort and time” by not re-deriving the same regulatory lessons from scratch.
The “we knew and did nothing” evidence
What makes the parallel more than a metaphor is that policymakers said this out loud, in public, years ago — and then largely didn’t act.
At the May 2023 Senate Judiciary Committee hearing on AI (with Sam Altman testifying), senators repeatedly invoked Congress’s failure to regulate social media or pass meaningful data privacy legislation as the mistake they didn’t want to repeat with AI. That’s a two-year-old admission of the exact failure mode being discussed today, from the people with the power to prevent it.
A former Twitter/Meta head of public policy, Sean Evins, wrote a first-person account (Fortune, Oct 2025) of watching this play out from inside both companies: misinformation, algorithmic bias, and polarization all identified early, all deprioritized in favor of speed, and all eventually “too big, too embedded, too essential” to meaningfully regulate by the time anyone tried. His framing of the current AI moment — “we’re moving fast, breaking things, and building the plane while flying it” — is essentially an eyewitness confirmation of the Sanders/Schneier thesis from the inside of the industry.
Where the parallel is being contested
Worth including for balance: not everyone agrees the “repeating the mistake” framing is accurate as of 2026. The American Consumer Institute (April 2026) argues the opposite — that AI-driven safety tooling (age-appropriate content filters, behavioral monitoring apps like Aura, AI-assisted moderation) is actually outpacing the slower legislative process, and that a wave of new 2026 state laws risks locking in requirements that are already obsolete relative to what the technology can self-regulate. It’s a useful counterpoint: the claim isn’t that regulation is impossible, but that hard-coded legal mandates move slower than either the harm or the fix.
Meanwhile, actual state-level AI legislation is arriving faster than it did for social media — Virginia’s minors’ social media law, California’s social media account-cancellation law, and California’s Transparency in Frontier AI Act all took effect January 1, 2026, alongside more than 20 other California AI laws spanning employment, healthcare, and education. A Cooley lawyer covering this space in April 2026 described lawmakers as explicitly drawing on the “parallel history” of social media regulation — sometimes usefully, sometimes not, since the underlying technologies aren’t identical.
The gap
Notably absent: a book-length treatment of this specific argument. Jonathan Haidt’s The Anxious Generation documents the underestimated-harm side of the social media story but doesn’t extend the thesis to LLMs. The parallel currently lives in papers, think-tank publications, and executive op-eds — which makes sense given how recent the LLM wave is, but also means the argument hasn’t yet had a full-length, synthesized treatment. That’s arguably the opening for exactly the kind of piece you’re writing.
Sources
- Sanders, N. & Schneier, B. (2024). Let’s not make the same mistakes with AI that we made with social media. Belfer/Berkman Klein Center, Harvard. Also in MIT Technology Review.
- Napoli, P. M. & Adi, S. (2025). On moving fast and breaking things.. again: social media’s lessons for generative AI governance. Information, Communication & Society.
- Governing AI Companionship: Lessons from Social Media Regulation. Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’26).
- Generative AI Regulation Can Learn From Social Media Regulation. arXiv:2412.11335 (2024).
- Evins, S. (2025, Oct 11). I had a front-row seat to the social media revolution… The same mistakes are happening in AI. Fortune.
- Yahoo Finance/AOL (2023, May 17). Senators say they failed to act on social media, won’t make same mistake with AI.
- The American Consumer Institute (2026, April). The Growing Gap Between Social Media Regulation and Technological Innovation.
- Privacy Daily (2026, April 30). Despite Parallels, Companies Must Track Differences to Comply with AI Chatbot Laws.