Internal documents from Microsoft have shed light on the company’s deep concerns regarding the use of media content in artificial intelligence (AI) training, describing the practice as “the largest theft of labour in human history.” The revelations emerged amid ongoing debates about the legal and ethical implications of AI companies using copyrighted materials, such as news articles, without explicit permission or payment.
The documents, released as part of the New York Times Co v OpenAI Inc lawsuit, reveal Microsoft executives warning of the risks posed by bypassing media paywalls to scrape large volumes of content for AI training. This activity, they argued, could severely undermine the viability of original news organisations, potentially triggering what has been termed an AI “doom loop,” where the depletion of quality source material perpetuates a cycle of declining content generation.
The concerns come as Australian lawmakers consider the regulatory environment surrounding AI training practices. Representatives from leading AI firms, including Jason Kwon of OpenAI, testified before a parliamentary inquiry, asserting that Australia’s current copyright framework acts as a barrier to investment and innovation in the AI sector. Kwon suggested that unless copyright laws are relaxed to permit broader use of media content without licensing, AI companies might withhold billions of dollars in investment.
Australia currently hosts data centers used primarily for AI inference—processing real-time queries using existing models. The more contentious issue lies with data centers dedicated to AI training, where vast datasets are ingested to develop and refine the AI models themselves. When media content is incorporated in this training stage, the AI product effectively internalises the style, knowledge, and intellectual property of the original creators to produce a commercial product, often without remuneration to the content owners. This phenomenon extends beyond news media to encompass other creative industries, including music, publishing, and film.
AI companies have defended their position on two key grounds. First, they argue that Australia’s copyright laws lack provisions similar to the “fair use” doctrine present in some jurisdictions, or specific exceptions for text and data mining. However, fair use’s applicability to AI training remains highly disputed in regions such as the United States. Second, they claim that negotiating licenses with numerous rights holders is impractical. Critics counter this by pointing to existing models like Australia’s News Media Bargaining Code, which coordinated collective agreements between digital platforms and media organisations, as well as the recently passed News Bargaining Initiative that targets additional digital services.
Observers note that the major digital platforms and AI companies have generally refrained from proactive engagement with content creators in favor of seeking legislative changes that would permit unfettered access to copyrighted materials. This approach contrasts with the Australian government’s stance that platforms using media content must pay for it, as demonstrated by prior regulatory actions involving Google and Facebook.
Experts warn that allowing AI companies to use copyrighted content without compensation sets a concerning precedent. The broader pattern has seen dominant digital platforms extract valuable content without negotiation, later imposing excessive charges and disclaiming responsibility for harmful societal impacts linked to their algorithms. They argue that holding AI firms accountable for authorised content use and potential negative consequences aligns with principles applied in other commercial markets, and that granting them exemption would be unjustifiable.
Rod Sims, former chair of the Australian Competition & Consumer Commission and current enterprise professor at the Melbourne Institute, encapsulates the sentiment by emphasizing that “AI exceptionalism is not an appropriate way forward,” advocating instead for regulatory frameworks that ensure fair remuneration and responsibility in the AI industry.