The recent settlement between Anthropic and major publishers and authors signals something troubling: the idea that large AI companies will simply buy their way out of copyright disputes is becoming normalized. This trend is being sold as inevitable. It deserves more skepticism than it is getting.
On the surface, these settlements look reasonable. Companies compensate rights holders. Everyone moves on. But the underlying message troubles me: if you have enough capital, copyright law becomes a negotiable licensing fee rather than an enforceable boundary.
That dynamic matters more than most observers seem willing to admit.
The copyright questions at stake in AI training are genuinely unsettled. Courts have never definitively ruled whether training large language models on copyrighted text without explicit permission constitutes fair use or infringement. These are foundational questions about how intellectual property operates in the digital age. They should be resolved through legal precedent, not through a series of ad hoc settlements that only wealthy corporations can afford to cut.
Consider what happens when settlements become the path of least resistance. Smaller AI companies lack the resources to negotiate billion-dollar deals with publishing houses and author coalitions. They face a choice: operate with legally uncertain training data, or don't operate at all. That's not a level playing field. That's regulatory capture by capital.
Meanwhile, rights holders face their own perverse incentive structure. If you believe a settlement is coming eventually, why litigate? Why risk unfavorable precedent in court when you can hold out for a payout? Settlements reward patience and leverage, not the actual strength of the underlying legal claims.
This is not to say rights holders are wrong to pursue compensation. Authors and publishers have legitimate interests in how their work is used and monetized. But those interests are best protected through clear legal rules, not through a bazaar where only the richest players can afford to negotiate.
The broader concern is about what kind of IP regime we're building. Intellectual property law is supposed to balance incentives for creation with the public's access to knowledge and culture. When that balance is struck through private settlements rather than statutory law or precedent, we get a system where the outcome depends on negotiating power rather than legal principle.
Recent IP disputes in other contexts reveal how fragile this reliance on settlements can be. Consider trademark disputes, where companies increasingly settle rather than test whether parody or comparative uses are legally protected. Consider patent disputes, where settlement agreements sometimes contain provisions that suppress legitimate innovation. The pattern is consistent: money replaces clarity.
The AI copyright cases are special because they involve foundational questions about how copyright applies to computational training. These are not edge cases. They will determine the legal landscape for decades. Kicking them into settlement agreements means future generations inherit a patchwork of private deals rather than coherent public law.
There is also the question of what gets left behind. When a company settles, the terms are often confidential. Other creators, other companies, and the public lack clarity about what was actually agreed to and why. Litigation produces precedent. Settlements produce only bills paid and silence.
I am not arguing that settlements are inherently wrong. Sometimes they represent good-faith compromises between parties with legitimate competing interests. But they should be the exception for difficult IP questions, not the expected path.
The skepticism deserves to run in multiple directions. We should question whether rights holders are using settlement leverage to extract rents that exceed what copyright law would actually protect. We should question whether tech companies are using their balance sheets to buy immunity from legal accountability. And we should question whether policymakers are comfortable allowing private actors to effectively rewrite IP law through negotiated deals.
The copyright-AI question will define intellectual property in the coming decades. That question is too important to resolve in confidential rooms between the world's richest companies and the world's largest publishing houses. It deserves a public answer.