Dossier mode
The Section 230 Playbook
The same investigation, restaged one beat at a time. Drive it with the arrow keys, space, or autoplay. Nothing is cut from the piece — long runs are split across frames. Read the full investigation or open the The Tech Right hub.
The Section 230 playbook.
30 years ago we called the internet 'wholly different' and gave platforms immunity, preempting state law. Zephyr Teachout says the AI industry is running the same move — and the 2025 record backs her up.
COI up front: drafted with Anthropic's model, so we grade our maker HARDER — and honestly, on these asks Anthropic is the counterexample (it opposed the moratorium, backed SB 53). The documented asks (preemption, a failed freeze, a Dec 2025 EO) are FACT; Teachout's 'preemption = back-door immunity' reading is PROBABLY TRUE.
We do NOT claim every AI firm wants total immunity or that a federal framework is inherently bad. The dominant 2025 legal ask, though, was exactly the escape-from-law move she names.
The precedent is real: Section 230 gave platforms broad immunity and preempted state law.
Section 230 of the 1996 CDA shields interactive-computer-service providers from liability for others' content, and courts read subsection (e) to preempt most state laws imposing such liability — carving the internet out of ordinary common law and state law. Scholars note the preemptive scope has been applied expansively. It's the clearest modern case of an industry placed in its own special legal category — the precedent Teachout says is being reached for again.
Today's ask #1: federal preemption of state AI laws — OpenAI requested it; a Dec 2025 EO pursues it.
OpenAI asked Congress in 2025 for federal preemption to escape a wave of state AI bills. On Dec 11, 2025 Trump issued an EO advancing preemption — 'a minimally burdensome national standard, not 50 discordant State ones' — directing AI czar David Sacks and Commerce to identify and challenge conflicting state AI laws and draft a federal framework. 36 bipartisan state AGs oppose a federal ban, and analysts say the executive branch likely can't preempt this way without Congress. Stripping states out is the current policy, not a prediction.
Today's ask #2: the moratorium — Congress twice tried to freeze all state AI regulation; the Senate stripped a 10-year version 99–1.
The nakedest 'wholly different regime' move was the moratorium: in 2025 congressional Republicans twice tried to attach a sweeping preemption of state AI laws — at one point a 10-year freeze — to must-pass bills, barring states from regulating AI at all while no federal regime existed. The Senate stripped the 10-year moratorium in a 99–1 vote (July 2025). It failed, but it's the clearest evidence of the goal Teachout names: not a rule, but exemption from the states' power to make rules.
Teachout's reading: preemption is a back-door immunity — it deletes the state tort and consumer-protection enforcement where accountability lives.
The interpretive core, graded as strong analysis. Most real accountability for corporate harm runs through state law — tort suits, state consumer-protection statutes, and AGs acting parens patriae. Preempting or freezing those doesn't just 'harmonize'; it removes the venues where injured people and enforcers can reach a company — the same shield Section 230 gave platforms. Whether framed as 'immunity' or 'a single national standard,' the effect is a preemption regime with a weak/captured federal backstop and little law with teeth. 'Total immunity from tort' is her characterization of the effect, not the companies' slogan.
The honest complication: the industry is split — and our own maker is the counterexample.
The preemption/moratorium push is driven hardest by a16z, OpenAI's DC operation, and the administration's David Sacks — but the industry isn't one voice. Anthropic (which makes the model that wrote this) opposed the 10-year moratorium — Amodei called it 'too blunt an instrument' leaving 'no ability for states to act, and no national policy as a backstop' — and endorsed California's SB 53 transparency law while the CTA lobbied against it. Reported straight, un-inflated: Anthropic still wants a federal framework and benefits from safety-first positioning, and backing disclosure is smaller than accepting tort exposure. But 'the AI companies' isn't uniform; the honest claim targets the dominant preemption push.
Her alternative, stated fairly: keep AI inside existing law; build any federal regulator within an existing agency to resist capture.
Presented as her argued position, its internal logic graded. Teachout: take AI harms seriously by using MORE law, not less — preserve state law and tort, strengthen consumer protection with real private rights of action, use antitrust against collusion, and rely on decentralized enforcement (AGs parens patriae) so federal capture isn't wholesale capture. Where a federal regulator is warranted, build it inside an existing agency (NRC/NNSA model), not a new single-purpose body easier to capture — by industry and by the 'post-law' EA ideology. The structural claim (decentralization hedges capture; preemption + a new regulator concentrates the risk it claims to manage) is sound governance.
The most powerful ask isn't for a rule — it's to be outside the rules.
The watched fight — what AI rules should say — is downstream of a quieter one: whether ordinary law applies at all. That's the tissue to The Wrong AI Debate (speculative risk crowding out present harm) and The AI-Safety Money ('safety' as an incumbency moat). A single federal regulator, especially a new one, is far easier to capture than 50 legislatures, hundreds of tort juries, and dozens of AGs — so decentralization is a hedge, and preemption-plus-immunity concentrates risk under a banner of managing it. Section 230 is the cautionary tale: a 'wholly different' carve-out, sold as nurturing a young industry, that hardened into a shield the biggest firms now guard fiercely.