Meta allegedly used shadow libraries, including LibGen, to train its Llama models without permission. The filings indicate that Mark Zuckerberg, was aware of and approved the use of pirated datasets.

Internal communications reveal that the decision to use LibGen was "escalated to MZ" (Mark Zuckerberg), who approved it despite concerns raised by Meta's AI executive team​.

Asking AI services: Please read the latest filings in the Kadrey vs Meta case and tell me what the new allegations are, what Mark Zuckerberg's role might have been, and what Meta should be doing to address those allegations and what they should have done to prevent this situation from occurring in the first place.

ChatGPT-4o’s answer:
New Allegations in Kadrey vs. Meta Case

The latest filings in the Kadrey vs. Meta case reveal serious new allegations, including:

1. Use of Pirated Datasets: Meta allegedly used shadow libraries, including LibGen, to train its Llama models without permission. These datasets contained copyrighted works from publishers such as Cengage, Macmillan, McGraw Hill, and Pearson​.

2. Stripping of Copyright Management Information (CMI): Meta is accused of removing CMI from these datasets to conceal copyright infringement. This stripping allegedly allowed Meta to evade detection and facilitated the illegal use of these works in training its AI models​​.

3. Torrenting and Seeding Pirated Data: Meta allegedly downloaded and distributed pirated files through torrenting, with internal communications showing employees' concerns about the legality of such practices. Despite these concerns, Meta reportedly continued the activity to build its AI models​.

4. Concealing Infringement in Llama’s Outputs: The complaint suggests that Meta fine-tuned Llama's outputs to avoid revealing the source of its training data, further obscuring its use of pirated materials​.

---

Mark Zuckerberg's Role

The filings indicate that Meta's CEO, Mark Zuckerberg, was aware of and approved the use of pirated datasets like LibGen. Internal communications reveal that the decision to use LibGen was "escalated to MZ" (Mark Zuckerberg), who approved it despite concerns raised by Meta's AI executive team​. Zuckerberg allegedly recognized the risks, including potential regulatory scrutiny and negative media coverage, but proceeded with the strategy​.

---

Recommendations for Meta

Addressing Allegations

1. Full Disclosure and Cooperation:

   • Meta should immediately cooperate with the court and provide all requested documents.

   • Conduct an independent investigation into the allegations, especially the involvement of senior leadership.

2. Settlement Discussions:

   • Engage in settlement talks to address the claims and avoid further litigation risks and reputational harm.

3. Enhanced Compliance:

   • Audit all datasets used in AI training to verify their legality.

   • Cease the use of pirated or questionable datasets.

4. Corrective Actions:

   • Publicly commit to ethical AI practices.

   • Offer restitution or licensing fees to affected copyright holders.

Preventive Measures

1. Data Governance and Licensing:

   • Implement a strict data governance framework to ensure datasets used for training are legally sourced and licensed.

   • Establish clear accountability for decisions about dataset acquisitions.

2. Training and Awareness:

   • Conduct mandatory training for employees, including leadership, on copyright laws and ethical AI practices.

3. Technological Safeguards:

   • Develop or acquire tools to screen datasets for copyright compliance before integrating them into AI training.

4. Oversight and Accountability:

   • Set up an independent oversight body to monitor compliance with legal and ethical standards.

   • Clearly define roles and responsibilities to avoid overreliance on top executives for critical compliance decisions.

---

By taking these steps, Meta can address current legal challenges and reduce the risk of future violations, ensuring greater compliance and trust in its AI development processes.

Text continues below the image»

Grok’s answer:
Based on the documents provided from the Kadrey vs. Meta Platforms, Inc. case, here's an overview:

New Allegations:

1. DMCA Violations:

   • Stripping of Copyright Management Information (CMI): Plaintiffs allege that Meta systematically removed copyright information from copyrighted works used in training its Llama AI models. This was done to prevent the AI from outputting copyright information, thereby facilitating and concealing widespread copyright infringement. Specific evidence includes:

     • Meta employee Nikolay Bashlykov admitting to removing "copyright" information from the Libgen dataset.

     • Documents showing Meta removed copyright paragraphs from documents, including science journal articles.

     • Confirmation from Meta's corporate representative about removing CMI to hide data sources used for training.

2. CDAFA Violations:

   • Unauthorized Access via Torrenting: There are claims that Meta accessed and distributed copyrighted data through torrent networks without permission, specifically:

     • Meta used torrenting to download and distribute pirated datasets like LibGen for training its AI models.

     • Meta's corporate representative admitted to uploading (seeding) pirated files, which supports the claim under the California Comprehensive Computer Data Access and Fraud Act (CDAFA).

3. Conformity with Evidence:

   • Updates to Reflect New Llama Models: The complaint seeks to include all versions of Meta's Llama models in the allegations, reflecting ongoing development since the original complaint was filed.

Mark Zuckerberg's Role:

• Approval of Pirated Dataset Use: According to new evidence, Mark Zuckerberg approved Meta's use of the LibGen dataset, which was known to be pirated, despite internal concerns about its legality. This approval was noted in internal memos where it was mentioned that after "escalation to MZ" (Mark Zuckerberg), Meta's AI team was given the go-ahead to use LibGen.

• Testimony on Piracy: During his deposition, Zuckerberg acknowledged that the activities described (like torrenting and using pirated data) would raise "lots of red flags" and seemed "like a bad thing," suggesting awareness of the ethical and legal implications.

What Meta Should Do to Address These Allegations:

• Legal Defense:

  • Motion to Dismiss: Meta could challenge the legal standing or the sufficiency of the evidence for the DMCA and CDAFA claims.

  • Evidence Presentation: Meta should prepare to counter these allegations with evidence that might show compliance with copyright laws or that the use of data was within fair use or other legal exemptions.

• Compliance and Remediation:

  • Review Data Practices: Immediately reassess data acquisition and handling practices to ensure compliance with copyright and data privacy laws.

  • Ethical Data Use: Implement stringent protocols for ethical data sourcing, including securing necessary licenses or permissions for dataset use.

• Transparency and Communication:

  • Public Statements: Clarify the company's position on data use in AI development, emphasizing ethical practices.

  • Internal Training: Conduct training for employees on intellectual property rights and data usage ethics.

Preventive Measures Meta Should Have Taken:

• Robust Legal Oversight:

  • Pre-emptive Legal Reviews: Before using any dataset, especially those with questionable origins, Meta should have conducted thorough legal reviews to ensure compliance with copyright laws.

• Data Sourcing Strategy:

  • Legally Sound Data Acquisition: Invest in or develop partnerships for legal data acquisition, avoiding reliance on pirated sources.

• Cultural Shift:

  • Ethical AI Development: Foster a corporate culture that prioritizes ethical considerations in AI development, with clear policies against using unauthorized data.

• Ongoing Compliance Monitoring:

  • Regular Audits: Establish regular audits of data usage in AI projects to ensure ongoing compliance with all relevant laws.

By addressing these points, Meta can mitigate current legal challenges and work towards preventing similar issues in the future.


Perplexity’s answer:
Based on the latest filings in the Kadrey vs Meta case, the plaintiffs are seeking to file a Third Amended Consolidated Complaint (TACC) with several new allegations against Meta Platforms, Inc. The key new allegations include:

1. Meta illegally stripped Copyright Management Information (CMI) from copyrighted works used to train its Llama models.

2. Meta participated in torrent sharing, both downloading and uploading pirated files containing plaintiffs' works.

3. Meta used multiple pirated datasets, including "LibGen," to train various versions of its Llama models.

4. Meta attempted to conceal its use of copyrighted material by removing metadata and using "supervised samples" when fine-tuning Llama.

While the filing does not specifically mention Mark Zuckerberg's role, as CEO of Meta, he could potentially be held responsible for overseeing the company's AI development strategies and ensuring compliance with copyright laws.To address these allegations, Meta should:

1. Immediately cease using any pirated or unauthorized datasets for training its AI models.

2. Conduct a thorough internal investigation into the use of copyrighted materials in AI training.

3. Implement strict protocols for vetting and obtaining proper licenses for all training data.

4. Cooperate fully with the legal proceedings and provide transparent information about its data usage practices.

5. Develop and implement robust ethical guidelines for AI development, emphasizing respect for intellectual property rights.

To have prevented this situation, Meta should have:

1. Established clear ethical guidelines for AI development from the outset.

2. Invested in creating or acquiring properly licensed datasets for AI training.

3. Implemented rigorous legal review processes for all data sources before use in AI training.

4. Developed technology to detect and exclude copyrighted material from training datasets without permission.

5. Fostered a corporate culture that prioritizes ethical AI development and respect for intellectual property rights.

By taking these steps, Meta could have avoided the legal challenges it now faces and set a positive example for ethical AI development in the industry.


Claude’s answer:
Based on the court filing, here are the key points:

New Allegations:

1. Meta systematically stripped Copyright Management Information (CMI) from copyrighted works used to train Llama, removing copyright paragraphs and metadata

2. Meta actively participated in torrenting pirated content - not just downloading but also "seeding" (uploading) files

3. Meta's practices extended beyond the initially disclosed Books3 dataset to multiple "shadow libraries" of pirated content

4. Senior executives, including Meta's head of Generative AI Ahmad Ah-Dahle, "cleared the path" for using these illicit sources despite internal concerns about legality

Regarding Mark Zuckerberg: While the filing doesn't directly mention Zuckerberg's involvement, there are references to recent statements he made about Meta's AI infrastructure investments. The filing cites an October 30, 2024 investor call where Zuckerberg stated Meta is training Llama 4 on a GPU cluster "bigger than anything else that I've seen reported for what others are doing." This suggests Zuckerberg was actively involved in and aware of the scale of Meta's AI training operations, though his direct knowledge of the data sourcing practices isn't addressed in this filing.

What Meta Should Do Now:

1. Immediately cease any ongoing use of unauthorized training data

2. Conduct a comprehensive audit of all AI training data sources

3. Implement robust data provenance tracking systems

4. Consider establishing a compensation framework for affected copyright holders

5. Cooperate fully with the legal proceedings and be transparent about past practices

6. Develop clear policies and procedures for legal data acquisition

7. Create an oversight committee for AI training data compliance

What Meta Should Have Done:

1. Established proper licensing agreements with content owners

2. Created clear policies prohibiting use of pirated materials

3. Implemented thorough vetting procedures for training data

4. Set up proper oversight mechanisms for data acquisition

5. Documented all data sources and usage rights

6. Taken internal legal concerns seriously rather than overriding them

7. Built systems to properly track and maintain copyright information

8. Prioritized legal compliance over rapid development

9. Invested in building relationships with legitimate content providers

The case highlights significant governance failures in Meta's AI development process, suggesting they prioritized speed and competitive advantage over proper data rights management and legal compliance.