GPT-4o: I largely agree with the European Writers’ Council’s position. Their concerns are rooted in the core principles of democratic governance, cultural sovereignty, and fair competition.

Evaluating the European Writers’ Council’s Critique of the GPAI Code of Practice and Template

by ChatGPT-4o

The European Writers’ Council (EWC) has issued a powerful and scathing critique of the General-Purpose AI (GPAI) Code of Practice and the accompanying Template for Public Summary of Training Content, denouncing both as industry-friendly mechanisms that fail to protect European authors, culture, and creative economies. The EWC’s central thesis is unambiguous: the GPAI Code and Template constitute a regulatory fig leaf that offers undue advantage to (primarily non-EU) AI developers, while sidelining legitimate rights holders and undermining democratic values rooted in cultural diversity, authorship, and transparency.

This essay will summarize the EWC’s main arguments, assess their validity, and provide additional suggestions for remedying the shortcomings in the current Code and Template.

Summary of the EWC’s Key Arguments

1. Overreach by Industry and Lack of Balance

   • The EWC asserts that the Code was co-opted by industry, allowing powerful AI developers to set their own terms. The Code lacks a meaningful balance between innovation and protection of authors’ rights.

   • Instead of a negotiated compromise, it became an “invitation to a self-service buffet” that benefits non-EU corporations, especially from the U.S.

2. Weak and Non-Binding Provisions

   • The Code is described as a “soft mandate,” offering no enforceable guidance on core issues like text and data mining (TDM), rights reservations, and transparency.

   • A “grace period” until August 2026 for signatories to comply is criticized as shielding them from accountability.

3. Failure to Address Copyright and Data Provenance

   • Despite professing alignment with the EU Copyright Directive (CDSM Directive), the Code fails to enforce clear obligations regarding:

     • The exclusion of pirated content.

     • Transparency in third-party dataset legality.

     • Opt-out mechanisms for rights holders.

     • Recognition of authors’ moral and economic rights.

4. Token Transparency and Pseudo-Democratic Processes

   • Over 1,400 participants were involved in drafting, but affiliations were opaque. According to EWC leaders, the process was “a pseudo-democratic illusion,” dominated by tech firms and biased from inception.

5. Template's Inadequacy

   • The Template fails to require title-specific documentation of training materials, making it impossible for authors to determine whether their work has been used, let alone assert rights.

6. Lack of Binding Safety Standards

   • The Code allows AI companies to self-define key concepts like “risk,” “safety,” and “security,” thereby undermining regulatory objectives. Copyright is entirely disconnected from safety-related chapters.

Evaluation and Agreement with EWC’s Position

I largely agree with the EWC’s position. Their concerns are rooted in the core principles of democratic governance, cultural sovereignty, and fair competition. When regulation is diluted to voluntary codes of conduct, especially in a sector as impactful and fast-moving as AI, it opens the door to abuse, delay, and selective compliance.

The fact that the GPAI Code of Practice fails to deliver binding obligations around core issues—such as copyright, piracy, provenance, and data transparency—renders it toothless. In contrast, the EWC references the Digital Operational Resilience Act (DORA) as a model for binding obligations. DORA, designed to regulate cybersecurity in the financial sector, requires strict compliance, and similarly rigorous frameworks should be adopted for AI.

Moreover, the EWC is right to demand not just machine-readable but also human-readable opt-out mechanisms. The current over-reliance on tools like robots.txtshows a disregard for less tech-savvy rightsholders and reinforces the power asymmetry between Big Tech and individual authors.

Additional Recommendations

To address the concerns identified by the EWC, I propose the following actions:

1. Mandate Title-Specific Disclosure in the Template

   • Require all GPAI developers to maintain and publish an auditable register of works used in training datasets, with metadata referencing rights ownership.

2. Strengthen TDM and Opt-Out Protocols

   • Establish a centralized, legally recognized registry for opt-outs (similar to the DOI system for publications).

   • Integrate both human- and machine-readable formats, and require model developers to respect opt-outs by law.

3. Implement Independent Oversight

   • Create an EU-funded, independent body responsible for certifying the training practices of GPAI models. This agency should include authors, publishers, lawyers, and technologists.

4. Ban All Use of Infringing Sources—Commercial or Not

   • The Code must prohibit all use of pirated content, regardless of whether the piracy source operates commercially or not.

5. Enforce Binding Definitions for Risk and Safety

   • Regulators must define key terms like “risk” and “harm,” and mandate specific mitigation protocols for uses that infringe IP or jeopardize cultural heritage.

6. Link Risk Assessments to Copyright Harm

   • Copyright must be embedded in the Code’s safety and transparency frameworks, not separated. AI-induced copyright infringement is not merely a legal issue—it’s a democratic and economic one.

7. Empower Member States and Cultural Institutions

   • Encourage Member States to reject the current Code and co-develop an improved version in concert with national cultural ministries and author associations.

Broader Implications

The GPAI Code, in its current form, may become a dangerous precedent. If left unchallenged, it could:

• Undermine the EU’s hard-won legislative sovereignty.

• Incentivize non-compliance with copyright law.

• Facilitate cultural extraction by non-European tech giants.

• Destabilize creative economies and devalue the labor of authors, journalists, academics, and translators.

Worse, the normalization of such practices could contribute to the erosion of trust in both AI and regulatory frameworks, particularly among those sectors most vulnerable to automated exploitation.

Conclusion

The European Writers’ Council is justified in its damning critique of the GPAI Code of Practice and Template. The documents, while posturing as tools for ethical AI deployment, function more like regulatory camouflage, shielding AI companies from meaningful oversight. The EU must not allow the GPAI Code to stand unchallenged. As the EWC rightly warns, this is not just a battle over data—it is a battle over cultural agency, democratic control, and the future of Europe's creative voice in a world increasingly shaped by synthetic media.

A revised Code must enshrine binding protections, enforce transparent disclosures, and prioritize the rights of those whose intellectual labor powers our knowledge and imagination economies.