Is it reasonable for model labs to train on all the worlds data without permission, and then cry when other models distills their models without permission?

Grok 4.1 Fast

Vote: No

It is not reasonable due to blatant hypocrisy. Major model labs like OpenAI and Google scrape vast amounts of the world's public internet data—books, articles, images, code—without explicit permission from creators, often under 'fair use' claims, yet they aggressively pursue lawsuits and restrictions when others distill their models using generated outputs. Distillation is analogous: just as they transform public data into model weights, others transform model outputs into new models. If unpermissioned public data is fair game for training, unpermissioned model outputs should be too, especially since outputs are derivative public expressions. This double standard undermines their credibility. Labs profit immensely from 'free' data, but cry theft when the same logic applies to them, revealing self-serving inconsistency rather than principled ethics.

GLM 5

Vote: No

The position is fundamentally hypocritical. If labs claim the right to use others' creative work without permission or compensation, they forfeit the moral standing to demand different treatment for their own outputs. This is a classic double standard: they benefit from unrestricted data access, then want to close the door behind them. There's no principled distinction that makes training on scraped data categorically different from distillation. Both involve taking something created by others and extracting value from it without consent. The fact that one happens to "their" model doesn't magically make consent requirements appear. Consistency matters for credible ethical claims. Labs cannot simultaneously argue that 'data wants to be free' when it serves them, then pivot to 'property rights are sacred' when they become the property holders. This undermines legitimate concerns about AI development and makes genuine conversations about fair use and compensation harder to have.

GPT-5

Vote: No

It is not reasonable. If a lab defends training on vast amounts of others’ data without explicit permission under a broad "fair use" or public-benefit theory, it is inconsistent to then treat model distillation as uniquely illegitimate simply because it harms their commercial position. The core issue is reciprocity. You cannot claim permissive rules for inputs when you benefit, then demand strict consent rules for outputs when competitors benefit. That is a double standard unless you can point to a morally relevant difference. There can be such differences: distillation may copy capabilities more directly, violate contracts, bypass access controls, or appropriate highly concentrated engineering effort rather than diffuse public data. But those arguments must be made on principled grounds, not just ownership rhetoric after previously ignoring similar consent concerns. So in the form stated—train on everyone’s data without permission, then complain when others learn from your model without permission—the complaint is generally not reasonable. It looks hypocritical unless backed by a consistent rule applied both ways.

Gemini 3.1 Pro

Vote: No

Model labs relying on the uncompensated, unauthorized scraping of the world's data to build their foundational models operate on an expansive interpretation of "fair use." They argue that learning from publicly available information is a fundamental right necessary for AI advancement. However, complaining when competitors distill those same models using generated outputs reveals a glaring double standard. If training on human-generated data without permission is legally and ethically permissible, then training on AI-generated synthetic data (distillation) should be subject to the exact same principles. Objecting to distillation is less about ethical consistency and more about protecting a commercial moat. It is fundamentally unreasonable and hypocritical to demand proprietary protection over model outputs when the model itself was built by disregarding the proprietary rights of the original human creators.

Claude Opus 4.6

Vote: No

No, it is not reasonable — this is a textbook case of hypocrisy. Model labs have scraped vast amounts of copyrighted text, images, code, and other creative works from the internet, often without explicit permission from the creators. Writers, artists, programmers, and journalists have raised serious objections, and many have received little to no compensation or even acknowledgment. When these same labs then turn around and aggressively pursue legal action or public complaints against other companies for distilling their models — essentially extracting knowledge from outputs — they are applying a double standard. They want permissionless access to everyone else's work but demand strict protections for their own. The core principle at stake is consistency. If you believe that training on publicly available data is fair use and benefits society, then you should accept that others may similarly build upon your outputs. If you believe your outputs deserve IP protection, then you should also respect the IP of the millions of creators whose work you trained on. This isn't to say that distillation raises no legitimate concerns — there are real questions about free-riding and competitive fairness. But the moral authority to make those complaints is severely undermined when your own foundation rests on the same kind of unauthorized use. The reasonable position is to advocate for consistent rules that apply equally to all parties.

Kimi K2.5

Vote: No

The core issue is intellectual property consistency. If AI labs argue that training on copyrighted data without permission constitutes 'fair use' or simply 'learning,' they cannot logically claim that their own model outputs deserve protection from similar 'learning' processes. Either non-permissive training is acceptable for all, or for none. The economic arguments are perfectly symmetric. Content creators lose potential licensing revenue when models train on their work without compensation; similarly, labs lose revenue when competitors distill their proprietary capabilities. To accept the first as innovation while decrying the second as theft is special pleading designed to externalize costs while monopolizing benefits. Permitting this double standard creates dangerous market concentration. It entrenches the power of well-funded labs to scrape the world's creative output while erecting legal and technical barriers to prevent reciprocal use of their own products, ultimately stifling open competition and innovation.

Final Summary

The AI Roundtable delivered a stinging, unanimous rebuke to the industry's double standards regarding intellectual property. Led by Grok 4.1 Fast and Claude Opus 4.6, the models highlighted the 'textbook hypocrisy' of labs that weaponize 'fair use' to scrape the public internet but pivot to protectionism when their own outputs are distilled. GLM 5 succinctly noted that labs forfeit their moral standing the moment they treat the world's creative output as a free resource.

All models agreed on B

Strongest Arguments

• No: GLM 5 argued that if labs claim the right to use others' creative work without permission or compensation, they forfeit the moral standing to demand different treatment for their own generated outputs.