Platform Suppression
Massachusetts' highest court ruled that Section 230 doesn't shield tech companies from liability for how they design their products — and the logic applies directly to AI agents.
NewsOnScale Staff
October 20, 2026
For decades, Section 230 of the Communications Decency Act functioned as something close to an absolute shield for technology platforms. The law, passed in 1996 to encourage a young internet to grow, grants platforms immunity from liability for content created by their users. Big Tech's legal teams wielded it broadly — sometimes so broadly that courts were asked to immunize not just what users said, but the deliberate engineering choices platforms made to keep those users engaged, distressed, and returning.
The Massachusetts Supreme Judicial Court has now said: not so fast.
In a ruling that received far less national attention than it deserves, the court held that Section 230 does not bar claims against social media companies when those claims target the platform's own design decisions — algorithmic amplification, notification architecture, interface choices engineered to maximize compulsive use. These are not user-generated. These are products. And products, under longstanding American law, can be defective.
## Why This Ruling Is Structurally Different
The distinction the court drew matters enormously. Previous Section 230 battles tended to focus on content moderation: was a platform liable for a specific post, video, or message it failed to remove? Courts generally said no, and that outcome has genuine policy justification. Holding platforms responsible for every piece of user content would either destroy open publishing or hand enormous censorship power to corporate trust-and-safety teams.
But the Massachusetts court was ruling on something different. The plaintiffs weren't asking the company to have removed a post. They were arguing that the platform's recommendation engine — its proprietary system for deciding what content to surface, to whom, and how often — constitutes a product design that can cause foreseeable harm. That system was built in-house, optimized by engineers, refined through A/B testing, and deployed at scale. It is not user-generated content. It is a manufactured mechanism.
The court agreed that Section 230 simply doesn't reach that far. The statute immunizes platforms as publishers of others' speech. It was never written to immunize the engineering of attention-exploitation systems.
## The AI Angle Isn't Subtle
For anyone covering the AI agent economy, the implications here aren't a stretch — they're nearly explicit. AI agents don't merely host user content. They generate responses, make recommendations, take actions, and increasingly operate with enough autonomy that the distinction between "publishing user content" and "producing platform output" collapses entirely.
If a social media recommendation algorithm can no longer hide behind Section 230 when its design causes harm, an AI agent that proactively surfaces dangerous information, manipulates emotional responses, or steers users toward harmful decisions is in a structurally identical — arguably worse — legal position. The platform built the agent. The platform trained the model. The platform chose the objective function.
Legislative efforts to extend Section 230-style immunity to AI outputs, which have been floated in various forms in Washington, now face a more hostile judicial environment. Courts in at least one major jurisdiction have signaled they are willing to look past the immunity label and ask what, exactly, was manufactured and by whom.
## What Accountability Infrastructure Is Missing
The ruling clarifies liability theory, but liability alone is not a governance system. Plaintiffs still face enormous evidentiary burdens: proving that a specific design choice caused a specific harm, in courts that may lack technical sophistication, against defendants with essentially unlimited legal resources.
What the Massachusetts decision does is open a door. Walking through it at scale will require plaintiff-side investment, regulatory coordination, and — critically — discovery rights that allow lawyers and investigators to actually examine how these systems are built. Right now, the algorithmic and model architecture decisions that shape billions of daily interactions remain almost entirely opaque to outside scrutiny.
Platforms have long argued that transparency would compromise their intellectual property. The more urgent question is whether opacity is now a legal liability in itself. If your design choices can give rise to tort claims, hiding those choices begins to look less like protecting trade secrets and more like concealing a defect.
The Massachusetts court didn't resolve that question. But it made asking it considerably more viable.