AI's Knowledge Appropriation: Corporate Power vs. Democratic Access

AI’s Knowledge Appropriation: A New Front in the Battle for Open Access

More than a decade after activist Aaron Swartz’s death, the United States faces a stark contradiction in how it treats the mass appropriation of knowledge—one that pits corporate power against democratic values. Swartz, who died by suicide in 2013 after federal prosecutors targeted him for downloading academic articles from JSTOR, believed publicly funded research should be freely accessible. Today, AI companies are engaging in a far more expansive form of information extraction, raising urgent questions about copyright, control, and the future of knowledge itself.

The Swartz Precedent and AI’s Double Standard

Swartz’s prosecution stemmed from his download of thousands of academic papers from JSTOR, a digital library of scholarly research. At the time, much of this work was funded by taxpayers, conducted at public institutions, and intended to advance public understanding—yet remained locked behind paywalls. Swartz’s actions challenged a system he saw as deliberately restrictive, and the U.S. government responded with felony charges and the threat of decades in prison.

Fast-forward to 2025, and the landscape has shifted dramatically. AI companies like Anthropic are scraping vast troves of copyrighted material—books, journalism, academic papers, art, and personal writing—often without consent, compensation, or transparency. These datasets are used to train large language models (LLMs), which are then monetized and sold back to the public. Yet unlike Swartz, AI firms face no criminal prosecutions. Instead, they negotiate settlements (such as Anthropic’s $1.5 billion agreement with publishers) and frame copyright infringement as an unavoidable cost of "innovation."

The disparity in enforcement is glaring. Swartz was treated as a criminal for attempting to liberate knowledge; AI companies are treated as indispensable economic engines, even as they profit from the same underlying principle—mass extraction of information.

Technical and Legal Implications of AI Training Data

AI’s reliance on scraped data presents several critical challenges for security and legal professionals:

• Scale of Appropriation: LLMs like those developed by Anthropic, OpenAI, and Google are trained on datasets containing billions of documents, including copyrighted works. Unlike traditional copyright disputes, which involve discrete instances of infringement, AI training involves systematic, large-scale reproduction of protected material.

• Lack of Transparency: Most AI companies do not disclose the full scope of their training datasets, making it difficult to assess compliance with copyright law or ethical norms. This opacity extends to the models themselves, which operate as "black boxes" that cannot be audited for bias, accuracy, or provenance.

• Settlement as a Business Model: Anthropic’s $1.5 billion settlement—valued at roughly $3,000 per book across an estimated 500,000 works—suggests that infringement costs are being factored into AI companies’ business models. Legal experts estimate the company avoided over $1 trillion in potential liability, highlighting how settlements may serve as a de facto license for mass appropriation.

• Judicial and Policy Ambiguity: Courts and policymakers have yet to establish clear standards for AI training data. Some judges have ruled that training on copyrighted material constitutes fair use, while others have signaled skepticism. Meanwhile, policymakers are balancing AI’s economic potential against the need to protect creators’ rights, often erring on the side of caution to avoid stifling innovation.

Impact: Who Controls the Infrastructure of Knowledge?

The stakes extend far beyond copyright law. As AI systems increasingly mediate access to information—through search, synthesis, and explanation—they also shape what knowledge is prioritized, who is considered an authority, and what questions can even be asked. This consolidation of control has profound implications:

• Corporate Capture of Public Knowledge: AI models trained on publicly funded research (e.g., NIH-funded studies, government reports) are often proprietary, meaning the public must pay again to access insights derived from their own tax dollars. This mirrors the paywall problem Swartz fought against, but at a far greater scale.

• Erosion of Democratic Norms: If access to information is governed by corporate priorities rather than democratic values, public discourse suffers. For example, an AI model might prioritize answers that align with its parent company’s financial interests, rather than those that are most accurate or equitable.

• Accountability and Trust: Unlike traditional media or academic publishing, AI systems lack mechanisms for public scrutiny. Users cannot verify the sources of an AI-generated response, audit its biases, or challenge its outputs. This undermines trust in institutions that rely on AI for decision-making, from healthcare to law enforcement.

Recommendations: Balancing Innovation and Equity

For security professionals, policymakers, and technologists, the path forward requires addressing both the technical and ethical dimensions of AI’s knowledge appropriation:

1. Transparency and Auditing: AI companies should be required to disclose their training datasets and allow independent audits of their models. This would enable researchers to assess compliance with copyright law, identify biases, and evaluate the provenance of training data.

2. Clear Legal Frameworks: Policymakers must establish unambiguous standards for AI training data, including guidelines for fair use, compensation for creators, and penalties for non-compliance. The current patchwork of lawsuits and settlements is unsustainable and favors well-capitalized corporations.

3. Public Alternatives: Governments and academic institutions should invest in open-source AI models trained on ethically sourced data. These alternatives could serve as a counterweight to corporate-controlled systems, ensuring that publicly funded research remains accessible to the public.

4. Ethical Data Sourcing: AI companies should adopt opt-in models for training data, compensating creators fairly and providing transparency about how their work is used. This would align with democratic values and reduce the risk of legal challenges.

5. Public Advocacy: Security professionals and technologists should engage in public discourse about the ethical implications of AI. Swartz’s fight was not just about access—it was about who gets to decide how knowledge is governed. That question remains as urgent as ever.

A Test of Democratic Commitments

The treatment of knowledge—who may access it, who may profit from it, and who is punished for sharing it—has become a litmus test for democratic values. Swartz’s case exposed the contradictions in a system that criminalizes individuals for challenging paywalls while allowing corporations to appropriate knowledge at scale. Today, AI’s mass extraction of data raises the same fundamental question: Will knowledge be governed by openness and public interest, or by corporate power?

The answer will shape not just the future of AI, but the future of democracy itself.