At 30, the Law That Built the Internet Is Being Asked to Unmake Itself

NewsOnScale Staff

August 26, 2026

When Congress passed Section 230 of the Communications Decency Act in 1996, the commercial internet had roughly 40 million users worldwide and the dominant platform was AOL. The law's 26 words — granting online services immunity from liability for third-party content — were written to protect a technology that looked nothing like what it became. This week, the Senate Commerce Committee marked that law's 30th anniversary with a hearing that made one thing unmistakably clear: the political will to change Section 230 is broad, but the agreement on how to change it is almost nonexistent.

That ambiguity is not a minor procedural footnote. It is the central fact of platform governance in 2025. And for anyone tracking how algorithmic systems shape public discourse, civic participation, and the emerging AI agent economy, the Senate's inability to move past symbolic friction is itself a story worth examining closely.

## What the Hearing Actually Revealed

The Commerce Committee hearing, whose transcript was published by Tech Policy Press, featured the familiar collision of grievances that has defined Section 230 debates for years. Conservatives argued platforms use immunity to censor conservative voices without consequence. Progressives argued the same immunity allows platforms to profit from radicalization, harassment, and health misinformation without accountability. Both camps want reform. They want nearly opposite things.

What cut through the noise, for those paying attention, was a quieter but more legally significant development happening simultaneously in state courts. The Massachusetts Supreme Judicial Court ruled this month that Section 230 does not automatically shield platforms from claims rooted in product design — meaning the architecture of a platform, its recommendation engine, its engagement loops, its notification systems — can be litigated as a design defect rather than dismissed as a content moderation question. That is a meaningful doctrinal shift. It moves the legal exposure from what platforms host to how they are built.

This distinction matters enormously. It means a platform cannot simply point to user-generated content and invoke immunity. If the product design itself foreseeably causes harm, the liability analysis changes. Public Knowledge's concurrent analysis argued this is precisely where Section 230's edges were always blurry — and where courts, absent congressional action, are beginning to draw their own lines.

## The AI Complication Nobody Wants to Resolve

Overlaid on all of this is a question that the 1996 Congress could not have anticipated: what happens when the content isn't generated by a third-party user at all, but by an AI system the platform itself built, trained, and deployed?

Commentary from multiple policy organizations this week argued, with reasonable consistency, that Section 230 was never designed to cover AI-generated outputs. The law's immunity applies to platforms acting as distributors of others' speech. When a platform's own model generates the speech, the platform is closer to a publisher — or a manufacturer — than a neutral conduit. That argument has not yet been tested definitively in federal court, but it is coming.

For the AI agent economy specifically, this is the fault line to watch. Agents that interact with users, generate recommendations, draft communications, or execute tasks on behalf of users are not passive hosts of third-party content. They are active participants in the informational environment. Extending Section 230 immunity to cover their outputs would represent a significant expansion of a law that is already straining under its original scope.

## The Accountability Gap in Plain Sight

What the anniversary hearings have exposed is not a law that failed, exactly, but a law that succeeded so completely at enabling platform growth that retrofitting accountability onto that infrastructure now feels structurally difficult. Platforms built entire business models inside the immunity perimeter. Dismantling it carries real economic and architectural disruption.

But the alternative — allowing that same immunity to quietly migrate into AI systems by default, without explicit legislative sanction — would be a policy choice made by inaction rather than deliberation. Congress has had 30 years with Section 230. It has had roughly three years with large-scale generative AI. The window for deliberate choices is narrowing, and the hearings this week suggest the institution is not yet moving at the speed the technology requires.