Anthropic's AI Gets Tangled in the Web of Copyright & Fair Use

Artificial Intelligence (AI) is reshaping numerous industries with its transformative capabilities, yet it also introduces complex legal issues, particularly concerning copyright. One prominent case illustrating these challenges is that of Anthropic, the developer of the AI "Claude." This case brought into sharp focus the debate over fair use in the realm of AI, as a San Francisco court ruled that while using legally acquired books for AI training is considered fair use, retaining pirated books amounted to copyright infringement. The ruling underscores the nuanced distinction between lawful acquisition and use of data and the implications of illegal data retention, which could have reverberations across the AI industry and beyond. For more detailed insights, the full case analysis is available.¹

Understanding the doctrine of 'fair use' is crucial for comprehending how copyright laws are being navigated within the AI industry. In the United States, fair use allows for limited use of copyrighted material without direct permission from the copyright holder, provided the usage is for criticism, research, or educational purposes. With AI, the application of fair use becomes particularly relevant as companies use vast amounts of data to train models. The key argument in the Anthropic case was whether using copyrighted works for AI training could be seen as transformative, thus falling under fair use. Legal debates continue as AI's transformative processes often blur the lines yet do not retroactively justify the illegal acquisition of data, a finding that places critical emphasis on the origins of training data. Explore this evolving legal landscape.¹

The Anthropic trial is a significant touchstone for the future of AI and copyright. It not only highlights the potential legal consequences AI companies face when engaging with pirated content but also sets a precedent that may influence upcoming cases. If Anthropic is found liable in its upcoming December trial, it could face substantial damages, driving home the message that ethical data sourcing is not just a legal requirement but a business necessity. This case may prompt AI companies to pay more for licensed data, potentially altering economic models within the industry. The ongoing legal scrutiny of such cases reflects the growing tension between innovation in AI and the traditional protections afforded by copyright laws. A comprehensive look at the implications of this decision can be found.¹

Anthropic's legal challenges have become a focal point in the ongoing debate over copyright infringement and fair use within the realm of artificial intelligence. The company, known for creating the AI model Claude, was recently embroiled in a lawsuit that has drawn significant attention due to its implications for the AI industry at large. The crux of the issue lies in Anthropic's dual use of copyrighted materials: While it successfully defended its practice of using legally acquired books for training its AI model under the doctrine of fair use, it was found guilty of retaining a large collection of pirated books. This ruling by a San Francisco court underscores a critical distinction in copyright law: the legitimacy of data acquisition is just as important as the subsequent use of that data..¹

The legal proceedings against Anthropic highlight a pivotal moment in delineating the boundaries of fair use for AI training. The court's decision to clear Anthropic of using copyrighted materials for AI training, provided they were legally obtained, stands as a landmark ruling with far‑reaching consequences. This aspect was deemed transformative enough to qualify as fair use, a legal doctrine in US copyright law that allows for limited use of copyrighted materials without explicit permission, especially in domains such as research and education. However, the retention of digitized copies of pirated books posed a clear violation of copyright laws, setting a precedent that could influence future cases and guide AI companies in their data acquisition strategies..¹

The concept of fair use within the context of AI training is becoming increasingly complex, particularly as legal frameworks struggle to keep pace with technological advancements. In a landmark case involving Anthropic, the creator of the AI model Claude, the court ruled that while using legally obtained books for AI training constituted fair use, retaining pirated books did not. This case exemplifies the nuanced discussions surrounding what constitutes fair use in AI model training. According to U.S. copyright law, fair use permits the limited use of copyrighted material without needing permission from the copyright owner. This doctrine applies to purposes like research, commentary, and educational use. Within AI, this means that if AI training can be shown to create something new and transformative, it might be considered fair use. However, the boundaries of this doctrine are still being defined by ongoing legal battles. [Read more](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

This evolving legal landscape is underlined by the court's ruling on Anthropic's case, which distinguishes between legal acquisition and illegal retention of data. While the court acknowledged the fair use of legally acquired books for AI training, it also emphasized that storing digitized copies of pirated books violated copyright law. This distinction has set a precedent, reinforcing that the legality of data acquisition is a critical component in determining fair use. The case also highlights the increasing attention on ethical data acquisition practices within the AI industry, as companies grapple with balancing the cost and legality of the data they use for training models.

The implications of these legal interpretations are significant for the tech industry. For AI companies, there is a growing need to adopt ethical and legal data acquisition practices, as the potential penalties for copyright infringement are substantial. For instance, Anthropic faces the risk of a class‑action lawsuit with possible damages up to $150,000 per work for copyright infringement related to pirated books. This demonstrates how the financial risks associated with non‑compliance can be a major deterrent for companies that might otherwise consider bypassing legal requirements in the name of innovation.

Furthermore, expert opinions suggest that the court's decision in Anthropic's case sets a crucial precedent for how AI training might be approached legally. The court compared the AI training process to human learning—absorbing information from books to synthesize new outputs—a transformative process arguably qualifying as fair use. However, this analogy may not apply universally due to the systematic nature of AI data processing, which differs significantly from human cognition. Thus, while the ruling supports AI innovation, it also raises questions about the extent to which AI training methods should mirror or diverge from human learning processes.

As legal battles continue, the case has sparked a wider debate on the ethical responsibilities of AI companies concerning data acquisition and the fair compensation of creators whose works are used. With AI systems' reliance on vast datasets, sourced both legally and illegally, public perception of AI and its role in society could be influenced by how transparently and ethically these companies handle copyrighted materials. Going forward, this transparency might be essential in fostering trust and acceptance in the broader societal context.

The case of Anthropic brings to light the intricate challenges of distinguishing between legally acquired data and pirated content when it comes to AI training. This complexity is especially pronounced in the domain of copyrighted materials. According to a ruling in San Francisco, using legally obtained books to train AI is classified under fair use, a doctrine that allows limited use of copyrighted material without obtaining explicit permission. However, retaining digitized copies of pirated books falls outside these boundaries, clearly infringing copyright laws. The distinction, while seeming straightforward, marks a critical evaluation point for AI developers, who must ensure that their data sources are lawful in an environment often blurred by digital proliferation. ¹

The repercussions of the Anthropic trial extend far beyond just the notion of legal vs. pirated content. AI companies, like Anthropic, must navigate complex legal frameworks that intersect with industry practices. In Anthropic’s case, while the court supported their use of copyrighted yet legally acquired books for AI model training, it marked a stark violation when dealing with pirated books. Such rulings reflect the current judicial understanding but also foreshadow more comprehensive guidelines, as governments and tech entities work towards clearer regulations. The industry is maturing under increased scrutiny, pushing for ethical compliance with copyright laws as a standard practice, rather than an ideal aspiration. ¹

The case highlights an important legal precedent in AI training, suggesting a nuanced interpretation of fair use. While AI models benefit from training on vast data sets, ensuring those data sets align with copyright laws is crucial. The court's decision implies a growing consensus that ethical sourcing of data should be a priority, turning away from any reliance on pirated content. Through this lens, the industry faces the challenge of curating high‑quality training data that complies with legal norms, potentially increasing operational costs as companies pay for licensing fees. Yet, this could also drive innovation in creating new data licensing models that balance legal compliance with the needs for expansive data. ¹

This case also raises broader questions of ethics and accountability in AI development. By recognizing both the legal and illegal pathways of data acquisition, the legal system underscores the importance of transparency and integrity. These principles are not only central to the credibility of AI developers but also critical in maintaining trust with the public and content creators. As AI technologies continue to evolve, adherence to copyright laws must remain at the forefront, acting as a guiding framework for ethical AI development practices. The lessons learned from Anthropic's situation could be instrumental in shaping future policies and industry standards. ¹

In recent news, Anthropic, the developer behind the AI technology Claude, was embroiled in a legal case concerning the possession of pirated books. Specifically, the company was held accountable for maintaining over 7 million pirated books between 2021 and 2023. Although Anthropic was cleared of any wrongdoing concerning the use of legally acquired books for AI training, a San Francisco court found them guilty of copyright infringement for keeping digitized copies of unauthorized books. A forthcoming trial in December will explore the extent of damages owed, potentially shaping ongoing legal debates about AI's interaction with copyrighted materials. More on these findings can be found [here](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The court's decision delineates a significant distinction in the realm of digital copyright law and AI technology. It stipulates that although leveraging legally obtained copyrighted works for AI model training can constitute fair use, the retention of pirated copies does not fall under the same legal protection. This distinction not only affects Anthropic but also has broader implications for other AI companies in similar legal battles concerning copyrighted content. This case could serve as a precedent, influencing how future claims around AI and copyright are adjudicated. For further information, see the article on [Actuia](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

This groundbreaking case also illuminates the tension within copyright law's fair use doctrine, particularly when applied to sophisticated AI models. While the court acknowledged the transformative nature of AI training that aligns with fair use for legal materials, it firmly opposed the creation of digital libraries of pirated content. The verdict underscores the importance of ethical data practices and could drive companies to pursue more legitimate avenues for data acquisition, despite the associated cost and effort. For a detailed report on the case, visit [Actuia](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The potential penalties for Anthropic, if the case escalates to a class‑action lawsuit, could amount to $150,000 per pirated work distributed, significantly impacting not just financial liabilities but also setting a legal precedent for the AI industry at large. As the AI community observes the unfolding developments, the case remains a touchstone in the ever‑evolving intersection of technology and intellectual property law. More detailed insights are provided in this [Actuia article](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The court's ruling against Anthropic in the retention of pirated books marks a significant moment in the legal landscape surrounding artificial intelligence and copyright law. While the company was cleared for its use of legally acquired books under the fair use doctrine, retaining pirated books is seen as a clear infringement of copyright law. This decision underscores the importance of securing permission and complying with copyright requirements when acquiring data for AI training. Should the trial in December result in a class‑action lawsuit, Anthropic could face severe financial penalties, up to $150,000 per work, if found liable. This potential outcome highlights the massive financial risks associated with infringing copyright laws in the context of AI development.

The legal implications stretch beyond punitive measures, influencing the broader AI industry's approach to data acquisition. This case serves as a stark reminder that innovation cannot come at the expense of compliance with intellectual property laws. For AI developers, the court's decision presents a mandate to align closely with legal norms, stressing the risk of legal action and financial liability if such norms are disregarded.

In establishing the illegality of storing pirated books, the court delineates a boundary critical for future AI development. The ruling prompts AI firms to reassess their data acquisition strategies, favoring ethical practices and potentially costly licensing agreements. This shift may spark innovation in data licensing models that facilitate legal compliance while enabling technological advancement. From a legal perspective, the case may serve as a benchmark for future litigation concerning copyright infringement in AI.

The ruling also intensifies the ongoing debate about what constitutes fair use in the age of AI. As courts continue to delineate the scopes of transformative use against undue copyright infringement, AI companies must navigate these developing legal interpretations carefully. Compliance with copyright laws will not only avert legal risks but might also bolster public trust by demonstrating commitment to ethical standards. Ultimately, the outcome of this case could influence legislative efforts to address the complexities of copyright in the evolving landscape of artificial intelligence.

In recent years, the intersection of artificial intelligence (AI) and copyright law has sparked a series of compelling legal battles, particularly regarding the use of copyrighted materials in AI training. One notable case involves Anthropic, the creator of the AI model Claude, which was recently found guilty by a San Francisco court of retaining unauthorized copies of books, yet was vindicated on charges related to using legally acquired books for training purposes. This ruling provides a significant precedent for understanding the boundaries of fair use in the realm of AI, where the transformative nature of AI learning is acknowledged as a valid expansion of the fair use doctrine. However, the retention of over 7 million pirated books by Anthropic highlights the legal risks associated with acquiring data unlawfully, pointing to a critical need for ethical data practices in AI development. You can read more about this case and its implications on [Actuia's website](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The verdict against Anthropic is part of a broader discussion in the AI community about how existing copyright laws apply to new technologies. The case underscores the complexities that arise when traditional copyright frameworks are applied to digital innovations. While courts are beginning to recognize the transformative potential of AI model training as a fair use, they are concurrently signaling a zero‑tolerance stance on the illegal acquisition of copyrighted works. This dual stance reflects the industry's shift toward more ethical data sourcing and the importance of transparency in how AI companies collect and use data. As these cases continue to surface, they shape the future of AI's legal landscape, emphasizing the balance between protecting copyright holders and fostering innovation within ethical boundaries. For further details on Anthropic's legal challenges and the broader implications for the AI field, refer to this [comprehensive article](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The ramifications of the Anthropic case extend beyond legal interpretations to potential shifts in business practices across the AI sector. With the looming possibility of penalties reaching up to $150,000 per work for infringed copyrights, AI companies are incentivized to overhaul their data collection methods, focusing more on obtaining explicit permissions and purchasing licenses for copyrighted content. This not only has financial implications for firms that may now face higher operational costs but also places smaller startups at a disadvantage when competing with larger, resource‑rich organizations. The case is influencing the global dialogue on copyright and technology, as countries may adopt diverse approaches to regulate AI, warranting harmonized international legislation that balances innovation with rights protection. Insights into how these dynamics will likely unfold are discussed extensively at [Actuia](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The court's recent ruling on Anthropic's case has drawn mixed reactions from legal experts, who view it as setting an important precedent in the intersection of copyright law and artificial intelligence (AI). The decision, which deemed the use of legally acquired books for AI training as fair use, was welcomed as a pivotal affirmation of the transformative potential AI models hold [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/). However, the court's firm stance against the storage of pirated books emphasizes the legal imperatives of acquiring training data ethically, raising significant implications for the AI industry [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

One prominent viewpoint suggests that the court's ruling aligns the activities of AI companies with broader ethical and legal standards, urging the need for more stringent data acquisition practices [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/). Many experts argue that this legal clarity may drive AI companies to invest more in securing lawful data, potentially elevating costs but also fostering greater trust within the digital economy. This transition is seen as imperative for maintaining the balance between innovation and intellectual property rights protection [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

Furthermore, discussions among legal analysts emphasize the analogy drawn by the court, comparing AI training to human learning—a notion that supports considering AI model training as transformative under fair use [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/). Such comparisons underscore the transformative applications of AI while highlighting the new challenges that this unprecedented scale of data utilization presents to existing copyright frameworks [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

Experts also express a cautious outlook regarding the court's decision, as the distinction between the inputs and outputs of AI training processes remains contentious [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/). The court's focus on the legality of the inputs rather than the produced outputs sets a significant precedent that demands careful navigation in future copyright litigations involving AI [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The implications of the ruling for the AI industry could prove profound, encouraging the development of new licensing models and elevating the standards for data ethics [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/). By reinforcing the boundaries of fair use, this legal development invites AI companies to innovate within a legally safe environment, ultimately strengthening the industry's legal and ethical foundations [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The Anthropic case has stirred a whirlwind of public opinions across various stakeholders in the tech and literary fields. For proponents of AI development, the ruling that affirmed the use of legally obtained books as fair use in AI training was a step forward for innovation. They believe this sets a positive precedent for the development of AI technologies, providing assurance that using copyrighted materials in this way is lawful as long as these materials are obtained legally. The decision is seen as a critical victory for companies investing in AI, signaling a supportive legal environment for technological growth and creativity. However, others express a different sentiment, concerned that this ruling might encourage companies to exploit copyrighted content without proper compensation to authors more vigorously [¹].

Conversely, there is considerable public backlash directed towards Anthropic’s handling of pirated books. Critics argue that the potential maximum penalty of $150,000 per pirated work underscores the severity of copyright infringement and the necessity for AI companies to commit to ethical data sourcing practices. This perspective highlights a growing call for accountability and transparency in how AI companies gather and utilize data. The forthcoming trial to determine damages is anticipated to further impact public discourse, emphasizing that AI’s expansive capabilities should not override ethical considerations and legal boundaries. As companies in the AI industry follow the case closely, the outcomes may guide future behaviors in handling copyright issues [¹].

The broader implications of the Anthropic case resonate through the entire tech sector as it navigates the challenges of balancing innovation with legality. This landmark ruling has sparked a polarizing dialogue among experts and the public alike about the ethics of data acquisition for AI. Advocates for technological progress juxtapose the necessity of access to vast data volumes against the importance of respecting intellectual property rights. As such, the case serves as a microcosm of a larger conflict within the AI industry regarding sustainable and responsible development practices. This duality encapsulates the challenging path forwards for AI development, where legal frameworks must evolve to effectively address new technological realities [¹].

The case has also triggered a profound reflection on the ethical responsibilities of AI companies. Critics argue that retaining pirated books was a blatant violation of copyright that casts a shadow over the entire AI sector. Questions arise about the transparency and ethics of AI data practices, catalyzing a reevaluation of how data should be sourced and used. The focus is now on whether companies will adhere to more stringent ethical standards voluntarily or through regulatory compulsion, fostering a legal landscape where intellectual property and innovation coexist harmoniously [¹].

The recent ruling in the Anthropic case, where the San Francisco court found the company guilty of retaining pirated books while clearing it of copyright infringement for using legally obtained ones, sets a critical precedent for the AI and copyright sectors. This judgment underscores the distinction between fair use in AI model training and illegal data acquisition practices. The court's decision points to a future where AI companies must navigate a complex legal landscape to ensure compliance, especially in securing data through legitimate channels. The onus will be on AI developers to adopt ethical practices in acquiring materials for training purposes, potentially leading to increased operational costs as they strive to avoid legal pitfalls associated with piracy. More information about this case can be found.¹

The implications of the Anthropic case extend into various facets of the legal and technological environment. Economically, large AI companies may better absorb the costs associated with securing legal training material, whereas smaller firms might struggle, leading to potential shifts in industry dynamics. The fair use doctrine, as applied in this context, might ease some financial burdens but also places emphasis on ethical sourcing. This underscores the necessity for transparent practices that align with both innovation goals and intellectual property rights. Such cases highlight that while AI innovation is vital, it should not come at the expense of creators' rights to their work. Details on how fair use was interpreted in this case can be viewed.¹

Socially, the ruling has brought to light the ethical considerations AI companies must navigate. Transparency in how data is sourced and the company's responsibility to uphold the legal rights of content creators is now under greater scrutiny. As the public becomes more aware of these practices, AI companies will be pressured to demonstrate ethical compliance, which will be crucial in gaining consumer trust. This is particularly significant as AI becomes increasingly integrated into various aspects of daily life. The public reaction to this case can be explored in more detail.¹

Politically, the Anthropic ruling could catalyze new legislation concerning AI and copyright. As governments grapple with the balance between promoting technological advancement and protecting intellectual property, this case may prompt clearer legal frameworks or regulatory reforms. The global nature of AI technology and data acquisition further complicates this, necessitating international collaboration to harmonize legal standards across borders. This may lead to a future where standardized international guidelines govern AI data practices and copyright protection. More insights into these potential political ripple effects are available.¹

In conclusion, the intricate journey of AI development intersects significantly with the evolving landscape of copyright laws. The Anthropic case epitomizes the complexities at this intersection, revealing a nuanced balance between innovation and legal compliance. In AI, the delineation of 'fair use' as it pertains to legally acquired data emphasizes the judiciary's approach towards fostering technological advancements. However, the retention of pirated content starkly contrasts with legal frameworks, as evidenced by Anthropic's upcoming trial, which serves as a critical reminder to all AI entities about the potential legal repercussions of copyright infringement. This dual verdict accentuates the necessity for AI developers to prioritize ethical and lawful data acquisition practices, ensuring respect for intellectual property while pursuing cutting‑edge technological solutions [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The verdict against Anthropic for retaining pirated books, while simultaneously being cleared for fair use in their AI training activities, has created a precedent in distinguishing between the legality of resources used for AI training and the resultant application of these resources. The ruling highlights that legally obtaining copyrighted works is crucial to align with fair use principles. For AI developers, this division necessitates a more rigorous approach to resource acquisition, potentially invoking new business models focused on acquiring rights to data sources. As the AI industry continues to evolve, developers must navigate this challenging terrain with careful consideration of both ethical implications and compliance with copyright laws [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

Moreover, the implications of this case extend beyond Anthropic itself, likely encouraging a broader shift within the AI sector towards more transparent and legally sound data sourcing methodologies. As AI's integration into various sectors intensifies, aligning business operations with legal statutes will not only mitigate risk but also bolster public trust and confidence in AI technologies. Companies, particularly those with fewer resources, may need to innovate in how they acquire and use data legally, possibly driving collaborations or the development of novel licensing agreements. Ultimately, this ruling may serve as a catalyst for a more legally conscious industry landscape [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).

The potential penalties faced by Anthropic, up to $150,000 per work for class‑action lawsuits, highlight the financial stakes involved in navigating copyright laws. This serves as a stark warning for AI companies about the severe ramifications of copyright infringement. It underscores the importance of investing in legal compliance early in the AI development process, which may involve considerable upfront costs but is likely to pay dividends in avoiding costly litigations and fostering sustainable growth. Moving forward, the AI industry is likely to see increased dialogue concerning fair use and licensing, aiming to streamline practices that protect the interests of content creators while enabling innovative technologies [1](https://www.actuia.com/en/news/claude‑at‑the‑helm‑anthropic‑found‑guilty‑of‑retaining‑pirated‑books‑but‑cleared‑on‑ai‑training/).