Arthur Sidney: When Intermediaries Become Participants
AI's shift from hosting content to actively shaping it means the real legal challenge isn't Section 230's survival but how to classify these systems at all, since that classification determines who's accountable when they fail.
Much of today's debate around Section 230 still assumes the internet of the late 1990s. The statute was written for platforms that hosted the speech of others, allowing them to moderate user content without assuming publisher liability and, in doing so, helping enable the modern internet. Artificial intelligence increasingly performs a different function. It does not simply host information. It helps produce and organize it. That shift raises a question that extends beyond liability: how should law classify AI systems?
But that is not the internet we are now building.
Too much of the current debate treats artificial intelligence systems as if they are simply larger or faster versions of the platforms Section 230 was designed to protect. That assumption is becoming harder to defend. AI systems are rapidly becoming part of the institutional infrastructure that underpins search, communication, education, commerce, healthcare, finance, software development, and public administration. They are not merely hosting speech. They are increasingly shaping how information is created, organized, prioritized, and ultimately used.
Artificial intelligence is transforming online services from passive hosts into active systems that generate, rank, summarize, recommend, retrieve, and increasingly act on information. AI search engines produce synthesized answers rather than lists of links. Enterprise copilots draft internal memos, generate code, and assist in procurement and hiring workflows. Customer-service chatbots resolve disputes autonomously. Retrieval-augmented systems curate and recombine external data in real time. Agentic systems are beginning to execute multi-step tasks with limited human input. These functions are qualitatively different from hosting third-party content. That shift matters because it places pressure on the legal categories that have governed internet platforms for nearly three decades.
Section 230 protects intermediaries from liability arising from the speech of others. But many AI systems operate in ways that are not cleanly about third-party speech. Their outputs reflect layered design choices: training data, model architecture, retrieval pipelines, ranking mechanisms, safety constraints, and generation strategies. Even when grounded in external sources, the resulting output is constructed, not merely transmitted.
Consider an AI search assistant. It retrieves content, weighs competing sources, resolves inconsistencies, and presents a single synthesized answer. That answer may be derived from third-party material, but it is not equivalent to displaying it. The system has materially shaped the information environment the user encounters. Enterprise copilots illustrate the same point. When an AI system assists in drafting legal analysis, recommending job candidates, generating financial forecasts, or supporting procurement decisions, it is participating in institutional decision-making. It is not simply passing along the speech of others; it is influencing how organizations interpret and act on information.
This is where the traditional intermediary framework begins to strain. For decades, internet law has relied on distinctions between publisher and speaker, platform and user, intermediary and author. AI systems increasingly blur these categories. They can simultaneously function as tools, editors, recommenders, and generators. The question is no longer just who created a piece of content, but how it was assembled, filtered, and presented.
Courts have already struggled to apply existing doctrine to recommendation systems, as illustrated by cases such as Gonzalez v. Google LLC, even before the widespread deployment of generative AI. Generative AI, retrieval-augmented systems, and agentic architectures intensify that challenge. As systems take a more active role in producing and organizing information, the boundary between hosting and participation becomes difficult to define.
The central issue is not whether AI systems should categorically receive immunity or liability. It is how they should be classified.
Classification determines far more than legal exposure. It defines which regulatory regimes apply, which agencies have jurisdiction, what compliance obligations attach, and how responsibility is distributed within firms and institutions. It also determines who must govern the system, who may intervene when it fails, and who remains accountable after deployment. These choices shape audit requirements, procurement standards, documentation practices, and expectations for human oversight. They also influence whether a system is evaluated as a communications intermediary, a product, a professional service, or a form of critical infrastructure.
In other words, classification is the legal foundation upon which governance and liability are built. Before lawmakers, regulators, courts, or organizations can assign responsibility, they must first determine what kind of system they are regulating.
If an AI system is treated primarily as an intermediary, governance will emphasize content moderation and speech protections. If it is treated as a product, attention shifts toward safety, testing, and defect liability. If it is treated as decision-support infrastructure, requirements may include auditability, explainability, and institutional accountability. If it is treated as an autonomous or semi-autonomous actor, entirely new frameworks may be required. These distinctions are not theoretical. They shape how systems are deployed in hiring, lending, healthcare, public benefits administration, and national security contexts. They determine who is accountable when systems fail—and what it means for them to fail in the first place.
Recent developments underscore the urgency of these questions. Beyond intermediary liability, copyright litigation is testing how generative models relate to underlying training data, while policymakers are advancing governance frameworks such as the NIST AI Risk Management Framework and the EU AI Act. Although these frameworks differ in scope and legal effect, they reflect a common premise: the obligations that attach to AI systems depend, in part, on how those systems are understood and classified.
Across these debates, the focus is shifting. The question is no longer only who published a piece of content, but what role an AI system played in producing, shaping, and operationalizing it. Section 230 helped build an internet defined by intermediaries. Artificial intelligence is creating systems that increasingly act as participants. That does not mean Section 230 is obsolete. But it does suggest that legal frameworks built around passive hosting may be poorly matched to technologies that actively structure information and influence decisions.
The debate ahead may not be whether Section 230 survives. It may be whether legal categories built around third-party speech remain adequate for systems that blur the line between host, editor, generator, and actor. Law cannot govern what it has not yet properly classified. Before responsibility can be assigned, law must determine what these systems are. The challenge is not simply one of liability. It is one of classification. And classification may prove to be one of the defining legal questions of the AI age.
Arthur D. Sidney is an attorney, policy advisor, adjunct professor, and founder of The Sidney Group PLLC, where he advises on AI governance, technology policy, institutional accountability, and regulatory strategy. He previously served for more than a decade on Capitol Hill, including as Chief of Staff and Chief Counsel to Congressman Hank Johnson, former Ranking Member of the House Judiciary Subcommittee on Courts, Intellectual Property, Artificial Intelligence, and the Internet. His writing has appeared in FedScoop, Nextgov, RealClearDefense, Techpoint Africa, and Broadband Breakfast. He is Director of Policy and Research for AI Safety Nigeria and serves as an Inner Council Member of the AI Sovereignty Network. This Expert Opinion is exclusive to Broadband Breakfast.
Broadband Breakfast accepts commentary from informed observers of the broadband scene. Please send pieces to commentary@breakfast.media. The views expressed in Expert Opinion pieces do not necessarily reflect the views of Broadband Breakfast and Breakfast Media LLC.
