AI copyright chaos: publishers demand fair pay and attribution

By early 2026, the legal landscape surrounding AI training data remained largely unsettled, with courts becoming the primary arbiters in the absence of definitive regulatory guidance. The Trump administration's cautious stance to 'let the courts decide' reflected widespread uncertainty about how traditional copyright frameworks apply to AI, especially as AI models increasingly use content without linking back or compensating original creators, a practice that disrupts the historical 'social contract' of the internet. Despite some large compensation deals, such as those involving Anthropic, only about 20% of publishers believed current models could sustain their businesses, underscoring a growing tension between AI companies and content owners over fair remuneration and ethical content use.

The landmark Anthropic Project Panama case crystallized key legal and ethical challenges in AI training data use. While a U.S. court ruled that Anthropic’s use of lawfully purchased copyrighted books for training constituted fair use due to the transformative nature of AI learning, the company simultaneously faced a $1.5 billion settlement for storing pirated copies, revealing a nuanced judicial distinction between training processes and unauthorized data retention. Internal documents exposing Anthropic’s covert plan to destructively scan up to two million books, physically destroying rare volumes to facilitate data acquisition, sparked public backlash and multiple lawsuits, highlighting the high financial and reputational risks AI firms face amid evolving copyright norms.

Globally, AI copyright litigation surged in 2026, exposing jurisdictional complexities and the urgent need for harmonized legal frameworks. High-profile lawsuits against AI giants like OpenAI, Meta, and Anthropic emerged not only in the U.S. but also in Europe, Japan, and India, where courts grappled with applying doctrines such as fair use and fair dealing to AI training. For instance, an Indian court ruled in OpenAI’s favor in the ANI Media case, emphasizing the specificity of the ruling while spotlighting gaps in Indian copyright law regarding AI. Meanwhile, Brazil explicitly rejected fair use for AI training, requiring compensation agreements, illustrating stark international divergences that complicate compliance for multinational AI developers.

At the heart of these disputes lies a fundamental tension between domestic copyright doctrines and international obligations under treaties like the Berne Convention and WIPO Copyright Treaty. U.S. courts have favored fair use in cases like Bartz v. Anthropic by emphasizing the transformative output of AI models, yet this approach often overlooks the reproduction inherent in the training process itself, creating friction with the Berne Convention’s strict three-step test for exceptions. The European Union’s copyright framework currently aligns more closely with international standards, while U.S. and Japanese laws lag behind, leaving AI developers to argue that their models learn statistical patterns rather than store copyrighted material. This disconnect underscores the pressing need for policymakers to craft balanced regimes that protect authors’ rights without stifling AI innovation.

By early 2026, publisher skepticism about the financial viability of selling content to AI models was palpable, with only about 20% believing such deals could sustain their businesses. Jim Lanzone highlighted that even seemingly large agreements, like those with Anthropic, fall short when broken down on a per-content basis, underscoring the urgent need for new, sustainable revenue models. Alongside compensation, there is a growing call for AI companies to adopt fair practices such as linking back to original sources, fostering an ecosystem that respects publishers’ rights before compensation debates become intractable.

By mid-2026, mounting legal pressures and legislative efforts catalyzed moves by tech giants like Google to proactively pay publishers for AI content use, aiming to preempt government mandates. This strategy leverages Google's vast advertising revenues—hundreds of billions annually—to potentially outmaneuver competitors like OpenAI, which face tighter margins due to high operational costs. Compensation models are evolving to focus on payments tied to direct citations or use of publisher content in AI responses, rather than indiscriminate payments for all crawled data, reflecting pragmatic responses to the massive scale of AI training datasets.

Publishers remain deeply concerned that AI platforms capture user attention and advertising revenue by withholding traffic from original content sources, thereby undermining traditional publisher revenue streams. Lawsuits from entities like The New York Times and coordinated legislative advocacy through groups such as the IAB reflect a concerted push to compel AI companies to compensate for what is perceived as unfair content appropriation and lost traffic. This tension highlights a core conflict between AI platforms’ business models and publishers’ economic sustainability.

Legislators and advocates like Laura Friedman emphasize the fundamental right of copyright holders to share in the value generated by AI systems trained on their materials, acknowledging multiple proposed compensation frameworks. While the current administration supports the use of publicly available content for AI training—as articulated in the president’s AI action plan launch speech—it has yet to endorse a definitive compensation model, leaving the debate open as the industry seeks balanced solutions.

By early 2026, a fundamental tension emerged between traditional internet attribution practices and the new realities of AI training data use. Jim Lanzone highlighted how the longstanding social contract of search engines linking back to original creators was upended by AI models that ingest content without direct attribution, raising profound ethical concerns about intellectual property rights. This shift not only challenges AI platforms but also exposes publishers to scrutiny, as many themselves fail to properly credit original sources, thereby denying creators recognition and potential revenue—a problem that calls for consistent attribution standards across both publishers and AI companies.

The ethical quandaries surrounding attribution intensified dramatically with revelations about Anthropic’s 'Project Panama,' which secretly destroyed up to 11,000 rare books daily by cutting off their spines to efficiently scan and convert them into locked PDFs for AI training. This destructive practice sparked outrage among literary advocates and copyright scholars, who condemned the irreversible loss of cultural artifacts designed to last centuries for the benefit of ephemeral AI models. Ed Newton-Rex encapsulated the controversy by pointing to a 'profound, systemic imbalance between trillion-dollar technology conglomerates and creative workers,' underscoring the ethical cost of prioritizing AI monopolies over authors’ rights and cultural preservation.

Anthropic’s aggressive data acquisition tactics are entangled in some of the largest copyright lawsuits in history, including a $1.5 billion settlement over the alleged piracy of seven million books, highlighting the unresolved legal and ethical tensions around fair use and compensation. Beyond legal battles, critics decry the monopolistic tendencies of leading AI firms, arguing that their centralized control over vast creative content—often used without explicit author consent—reflects an unethical philosophy of gatekeeping. This controversy not only questions the legitimacy of AI training methods but also calls into focus the broader implications of how attribution and compensation are managed in an industry rapidly reshaping the creative landscape.

Publishers remain skeptical about monetizing their content through AI licensing, with only about 20% confident they can earn sufficient revenue this way, underscoring the need for ethical attribution practices that go beyond mere compensation. Jim Lanzone emphasized that doing the right thing involves linking back to original sources to benefit both users and publishers, reinforcing that ethical content use is not just a legal issue but a social contract. This perspective advocates for a balanced ecosystem where AI companies and publishers alike uphold transparency and respect for creators, fostering a fairer digital environment amid rapid technological change.

The European Union has taken a pioneering role in crafting a regulatory framework that balances AI innovation with copyright protections, exemplified by its text-and-data mining exception modeled after the historic Statute of Anne. This exception allows AI to learn from lawfully accessible material without infringing on creators’ rights, focusing on whether outputs duplicate existing work rather than restricting inputs, a stance Brian Williamson argues is crucial for Europe to maintain its AI competitiveness despite infrastructural challenges. The EU AI Act further enforces transparency by requiring AI providers to disclose training data summaries and respect copyright opt-outs, applying these obligations extraterritorially to any model offered within the EU, thereby preventing circumvention through offshore training.

Despite the EU’s advances, global governance of AI training data remains fragmented due to divergent national copyright exceptions and interpretations under international treaties like the Berne Convention. While Article 9 grants authors exclusive reproduction rights, its allowance for national exceptions under a strict three-step test results in significant variation, with the United States’ broad fair use doctrine—illustrated in Bartz v. Anthropic—permitting large-scale commercial AI training that arguably conflicts with international obligations. This disconnect underscores the urgent need for new international agreements to harmonize protections, as current frameworks risk undermining authors’ rights by enabling extensive copying without consent or compensation.

Emerging legal disputes, such as the ANI Media v. OpenAI case in India, highlight the complex challenges of applying existing copyright laws to AI training, particularly regarding the scope of 'fair dealing' provisions and cross-border jurisdiction. OpenAI’s argument that Indian courts lack jurisdiction due to offshore training activities exemplifies the enforcement difficulties in a globally distributed AI ecosystem. This case has evolved into a broader policy debate involving diverse stakeholders—from the Indian Music Industry to technology startups—signaling a pressing need for comprehensive regulatory frameworks that reconcile innovation incentives with creators’ rights in the AI era.

Calls for compensation models are gaining momentum, with advocates like Laura Friedman emphasizing that copyright owners should share in the value generated when AI systems train on their materials. While the current U.S. administration supports the use of publicly available data for AI training, it has yet to endorse definitive compensation frameworks, reflecting ongoing tensions between fostering innovation and ensuring fair remuneration. This debate underscores the broader regulatory challenge of aligning domestic policies with international standards to create equitable and sustainable AI governance.

By mid-2026, AI companies like Anthropic have been aggressively expanding their training datasets through large-scale acquisitions of copyrighted books, often conducted discreetly to avoid public backlash, as exemplified by Anthropic’s Project Panama which aimed to scan all books worldwide. This competitive fervor is mirrored in the surge of rare book orders by entities such as Zoom Books, underscoring a broader market trend where comprehensive and diverse data collections are becoming a critical commercial asset driving AI innovation.

Legal rulings are increasingly shaping how AI firms navigate the complex terrain of copyright in training data use. A notable example is the judicial decision that Anthropic’s use of copyrighted books constituted fair use due to prior purchase, setting a precedent that influences corporate strategies and emboldens companies to leverage purchased content as a legal foothold. This evolving legal landscape not only affects data acquisition tactics but also impacts competitive dynamics by delineating the boundaries within which AI innovation can proceed.