Section 230’s shield wavers as courts eye chatbots

Section 230, the U.S. law that insulated early web platforms from a flood of lawsuits, may not reliably protect generative AI systems like ChatGPT. That emerging consensus among legal scholars — and recent comments from the statute’s co-author — is raising the stakes for how governments treat AI models that generate, rather than merely host, content.

In remarks at a 2023 briefing hosted by the Electronic Frontier Foundation, Senator Ron Wyden, who co-wrote Section 230, stressed that the law was designed to shield services from liability for user-generated content, not material they create themselves. He pointed to cases like Lemmon v. Snap, where Snapchat’s “speed filter” was treated as the company’s own product, outside 230’s protection.(wyden.senate.gov) That same logic, he and many legal experts suggest, could leave AI developers exposed when a model fabricates defamation, unlawful advice, or other harmful material.

The distinction sounds technical but cuts straight to the business model of generative AI. For social networks or forums, Section 230’s 26 words generally shift liability to the person who posts. For large language models, the controversial output is often an original synthesis composed by the system in response to a user prompt. As Matt Perault argued in Lawfare, that makes products like ChatGPT look less like message boards and more like speakers or publishers in their own right — a category Section 230 was never meant to protect.(lawfaremedia.org)

At the same time, lawmakers are already moving to narrow the law’s scope. A bipartisan push led by Senators Dick Durbin and Lindsey Graham would sunset Section 230 within two years, forcing Congress to rewrite the rules for online liability.(theverge.com) If that clock runs out before courts clarify how 230 applies to generative AI, companies building and hosting chatbots could find themselves in an especially precarious position.

The practical risks are multiplying. Generative models have already been accused of producing false statements about individuals, generating copyrighted or trade-secret–like text, and enabling scams or illegal instruction. Legal scholarship from the University of Chicago Business Law Review notes that AI-generated content “blurs the traditional distinction” between passive hosts and active publishers — and warns that, without a clear safe harbor, startups may struggle to attract investment or afford insurance.(businesslawreview.uchicago.edu)

So far, the U.S. Supreme Court has sidestepped the question. In Gonzalez v. Google and its companion case, Twitter v. Taamneh, the justices resolved terrorism-related claims on other grounds, leaving Section 230’s core interpretation largely untouched.(en.wikipedia.org) But these cases focused on recommendation algorithms for third-party videos and posts — not on systems that compose the speech themselves.

That gap is sending regulators elsewhere in the world down a different path. The European Union’s AI Act, finalized in late 2023, treats general-purpose and generative AI models as distinct products with their own transparency, safety, and documentation obligations, rather than trying to fit them under a US-style intermediary shield.(arxiv.org) Future conflicts between EU-style product liability for AI and a more speech-focused U.S. regime could fracture how global AI services are built and deployed.

Inside Washington, the politics are far from settled. At an Axios AI policy event, Senator Ted Cruz recently predicted that AI companies will “probably” end up covered by Section 230, even as he questioned whether they should be.(axios.com) Other lawmakers, including Wyden, warn that eroding liability protections risks pushing platforms toward over-removal and chilling online speech — especially for smaller services that lack the resources of Big Tech.(wyden.senate.gov)

For the AI industry, the stakes extend beyond courtroom losses. If courts decide that key parts of generative systems fall outside Section 230, providers could be forced to overhaul safety architectures, implement aggressive pre- and post-deployment filtering, and negotiate new contractual risk-sharing with enterprise customers. Some scholars argue that such pressure could ultimately drive more responsible design. Others fear it could cement the dominance of a handful of well-capitalized firms and limit open-source experimentation.

What is clear is that the comfortable legal story of the early web — platforms as neutral conduits for other people’s speech — no longer maps cleanly onto systems that fabricate sentences, images, or code on demand. As Section 230 turns 30 under political fire and generative AI powers ahead, courts and lawmakers are being pushed to answer a deceptively simple question: when a chatbot speaks, who is legally talking?