How "Original" is Generative AI: Approaching the Copyright Threshold with Fingerprints - A New Negotiation Technique to Turn "Similarity" into Value

1. What Happened: Visualized "Degree of Similarity"

The UK's 'The Guardian' introduced Vermillio on October 18 (US Eastern Time) as a platform that visualizes "how much generative AI tools rely on copyrighted works." Vermillio creates a "neural fingerprint" for each copyrighted work and matches it with AI outputs to calculate the degree of similarity. A video generated by Google Veo3 with a prompt reminiscent of Doctor Who showed about 80% similarity, while a video by OpenAI Sora showed about 87% similarity. Strong matches were also observed with Bond (James Bond), Jurassic Park, and Frozen.The Guardian

Vermillio advocates for "TraceID," a rights management platform encompassing IP/NIL (Name and Image Rights) protection, misuse tracking, and licensing proposals. Its distinctive feature is that it promises to "not just detect but monetize."vermill.io

2. The Perspective of Generative AI: Distance to Fair Use and "Third-Party Measurement"

According to the article, Google emphasized its policy against intellectual property infringement while stating, "We do not comment on the results of third-party tools." OpenAI explained that learning from publicly available data aligns with the US fair use doctrine. YouTube indicates in its terms that content may be used for machine learning. While these are criticized for "lack of transparency," they also represent survival strategies within varying copyright systems across jurisdictions.The Guardian

3. The Court Atmosphere: Anthropic's $1.5 Billion Settlement Reflects "Market Sense"

Coinciding with the timing, US-based Anthropic received preliminary approval for a $1.5 billion settlement in a class-action lawsuit over learning from pirated books. Reports indicate an expected payment of about $3,000 per book and the destruction of the data in question. Rights holders view this as groundbreaking, increasing pressure on AI companies to "create a licensing market."AP News

4. Industry Friction: Actors and Creators Push Back

The UK's actor union Equity declared it would not hesitate to take "large-scale direct action" against unauthorized AI use of likenesses and voices. The current situation, where generative AI's "live-action level" synthesis erodes trust on set, is evident.The Guardian

5. Reactions on Social Media: A Three-Way Battle of Welcome, Caution, and Skepticism

• Welcoming Side (Rights Holder Groups): The American Association of Publishers (AAP) positively evaluated the Anthropic settlement, calling it a step towards forming a licensing market. The Writers Guild also posted multiple threads and procedural guides, supporting the realization of rights enforcement.AAP

• Concern at the Scene (Performers): Support gathered for Equity's account and statements, visualizing the anger over actors' "digital duplication." Criticism of specific cases and hashtag movements spread.Equity

• Skeptical Side (Tech Community): Questions arose about the validity of metrics, such as "What does '% match' mean?" and "What about the reproducibility and false positive rate of fingerprinting?" Estimating similarity in images and videos involves various academic indicators, and conclusions can vary depending on the design, such as model fingerprints and Siamese networks for "explainable similarity."Proceedings of Machine Learning Research
  
  

6. Understanding the Technical "% Match": Three Points of Caution

(1) What is Being Compared
Whether it's frame-level features, composition, texture, or character style, academic layers differ, such as "latent fingerprints of generative models," "material copy detection," and "semantic similarity." Different metrics mean different percentages.Proceedings of Machine Learning Research

(2) Transparency of Detection
Opaque scores are weak in litigation and negotiation. Verifiable evidence trails on blockchain and interactions with public benchmarks become important.SpringerLink

(3) False Detections and "Cultural Symbols"
Cultural symbols like a blue phone box → "Doctor Who-ness" can easily mix specific works with abstract tropes. Explanations accompanied by process visualization (such as heat maps) are essential.MDPI

7. Still, "Visualization" Works: Reducing Negotiation Costs

Vermillio is not just about "exposure," but it streamlines the process from detection → evidence creation → licensing proposal. This aligns with digital-era rights enforcement, where "first, make the other party aware of your existence" is key. It can serve as a foundation for creating a "place of consent" among rights holders, platforms, and AI companies.vermill.io

8. Future Practices: Three Actions

• Contract and Prompt Regulation: Clearly state "verification, deletion, and regeneration obligations for AI-generated content" and prompt prohibitions for "famous IP-like" in production and distribution contracts.

• Mandatory Audit Logs: Require AI vendors to provide data provenance, integrate blocklists, and accept third-party audits.

• "Proactive Measures" by Rights Holders: Not only introduce detection tools but also establish licensing menus and clearly display pricing tables to achieve "pay if you want to use."

9. The Current State of Rule-Making

In the UK, artists continue to oppose the consideration of "AI exceptions" in copyright law. While corporate self-regulation has its limits, excessive regulation can stifle innovation. Therefore, it is crucial to advance both transparency (data provenance and explainability of detection methods) and fair compensation (settlement and comprehensive license market formation).The Guardian

Appendix: Key Facts from This Case

• In Vermillio's verification, Veo3 videos in the style of Doctor Who showed about 80% similarity, while Sora videos showed about 87% similarity. High similarity was also observed with Bond and others.The Guardian

• Google refrained from commenting on third-party measurements, while OpenAI asserted alignment with fair use.The Guardian

• Anthropic's $1.5 billion settlement has received preliminary approval, with a reported level of about $3,000 per work.AP News

• The actors' union Equity warns of "large-scale direct action."The Guardian