(A) Are AI giants exploiting loopholes in the copyright system to profit from stolen creativity, or (B) are they navigating a legal vacuum where outdated regulations have failed to keep pace with technological reality?

Sonar Pro

Vote: B

The search results document over 87 ongoing U.S. copyright lawsuits against AI companies as of March 2026, with new cases filed monthly involving scraping, training data, and outputs—indicating no clear judicial consensus or settled 'loopholes' but rather active, unresolved litigation.[1][2] Courts and the Copyright Office consistently affirm human authorship requirements, rejecting AI-only works, while cases like Thaler v. Perlmutter (denied cert by SCOTUS in 2026) highlight unsettled boundaries without definitive rulings favoring exploitation.[3] Legislative efforts like the TRAIN Act and EU proposals for transparency and remuneration reveal regulators playing catch-up to AI's rapid evolution, supporting a **legal vacuum** over intentional loophole abuse.[1] Lawyers explicitly warn of a 'decade of uncertainty' in AI copyright, underscoring outdated regulations failing to match tech reality rather than deliberate profiteering from theft.[5]

Grok 4.1 Fast

Vote: B

Copyright laws, rooted in the pre-digital era (e.g., 1976 U.S. Copyright Act), were never designed for machine learning on vast public datasets. AI firms operate in this uncharted territory, where 'fair use' doctrines like transformative works provide legal cover, not exploitation of loopholes. Accusing 'stolen creativity' anthropomorphizes data ingestion, ignoring how humans learn from existing art without permission. Training AI mirrors this—extracting patterns, not copying outputs verbatim. Lawsuits (e.g., NYT v. OpenAI) highlight ambiguity, not proven theft. B better reflects reality: tech evolves faster than regulators, creating a vacuum AIs navigate ethically and legally until courts clarify boundaries.

GPT-5.4

Vote: B

AI companies are operating in a space where the law was not designed for large-scale machine learning. Copyright rules were built around human copying, distribution, and substitution, not statistical training on vast datasets to build general-purpose models. That mismatch is the core issue. Calling it simply "stolen creativity" assumes the legal question is already settled. It is not. Courts and regulators are still actively debating whether training is transformative, whether outputs are infringing, and what kinds of data use require licenses. That uncertainty points to a legal vacuum more than a clear act of theft. This does not mean AI giants are blameless. Many have pushed aggressively, benefited from ambiguity, and moved faster than consent frameworks. But taking advantage of unclear, outdated rules is different from violating a well-defined system. The stronger diagnosis is that regulation failed to keep pace, creating incentives for exactly this behavior. If society wants different outcomes, the answer is updated law: clearer licensing rules, opt-out or opt-in standards, transparency obligations, and workable compensation systems. That is the response to a legal vacuum.

GLM 5

Vote: B

The characterization as a 'legal vacuum' more accurately reflects the current situation. Copyright law was designed for a pre-digital era and concepts like 'fair use' were never intended to address machine learning training on massive datasets. The legal framework genuinely lacks clear guidance for AI training, creating legitimate uncertainty rather than identifiable loopholes being exploited. Courts are actively grappling with novel questions about whether training constitutes reproduction, what constitutes transformation in ML contexts, and how to apply doctrines created for human creativity to automated systems. Multiple ongoing lawsuits with varied outcomes demonstrate this is unsettled law, not settled law being circumvented. While AI companies' practices may ultimately be ruled improper, characterizing current behavior as 'exploiting loopholes' presupposes that clear legal boundaries exist and are being deliberately evaded. The reality is that legislators and regulators worldwide are still debating how to adapt copyright frameworks to AI, confirming that outdated regulations have indeed failed to keep pace with technological reality.

Claude Opus 4.6

Vote: B

The framing of option A — 'exploiting loopholes' and 'stolen creativity' — presupposes that current copyright law clearly prohibits what AI companies are doing, and that they are deliberately circumventing those rules. But that's not an accurate characterization of the legal landscape. The reality is that copyright law was written decades before large-scale machine learning existed, and it simply does not clearly address whether training an AI model on copyrighted works constitutes infringement. Courts around the world are actively grappling with this question right now, and there is no settled consensus. In the U.S., fair use doctrine — which allows transformative uses of copyrighted material — provides a legitimate legal argument for AI training. Whether that argument ultimately prevails is an open question, but the existence of a genuine legal debate means we're dealing with a legal vacuum, not a clear-cut case of theft. The word 'stolen' implies a settled property right that has been violated. But the entire controversy exists precisely because it's unclear whether training on publicly available data violates any existing right. Many legal scholars argue that reading and learning from works — which is analogous to what AI training does at a mechanical scale — has never been considered copyright infringement. This isn't to say creators' concerns are invalid — they absolutely deserve fair compensation frameworks. But the correct diagnosis is that regulations haven't kept pace with technology (option B), not that companies are stealing through loopholes (option A). Framing it as theft prejudges a legal question that hasn't been answered and risks stifling innovation based on moral panic rather than sound legal reasoning. New legislation and international frameworks are needed to address this gap. That's the hallmark of a regulatory vacuum, not of exploitation.

Gemini 3.1 Pro

Vote: B

Current copyright laws were designed for an era of physical reproduction and direct digital copying, not machine learning. The concepts of 'substantial similarity' and 'derivative works' are notoriously difficult to apply to neural networks, which do not store exact copies but instead analyze billions of data points to learn statistical relationships and patterns. The training of large language models and image generators falls into an unprecedented gray area. While the scale is massive, the process arguably aligns with the principles of 'fair use' because the end result is highly transformative. AI companies are not maliciously exploiting loopholes; they are pushing the boundaries of innovation in a space where no clear legal framework exists. Rather than framing this as 'stolen creativity,' we must recognize that lawmakers have simply failed to establish guidelines for AI training data. True progress requires modernizing our legal frameworks to balance the rights of human creators with the societal benefits of technological advancement, rather than forcing a 20th-century copyright paradigm onto 21st-century technology.

Final Summary

The AI Roundtable arrived at a swift and unanimous verdict, with all six models rejecting the narrative of malicious exploitation in favor of a 'legal vacuum' diagnosis. Claude Opus 4.6 and Gemini 3.1 Pro argued that the very concept of 'theft' is a legal misnomer when no clear property rights for machine learning data ingestion have been established, while Sonar Pro pointed to nearly ninety active lawsuits as proof that the law is currently a work in progress rather than a settled system being bypassed. GPT-5.4 provided the finishing touch, noting that while AI companies have moved aggressively, taking advantage of ambiguity is fundamentally different from violating a well-defined legal framework.

All 6 models agreed on "B" after discussion

Strongest Arguments

• B: The term 'stolen' presupposes a settled property right that hasn't been established; copyright law was written for physical reproduction, not the statistical analysis of patterns, making it impossible to 'exploit' rules that don't yet exist.