AI Is Making Intellectual Property’s Flaws Impossible to Ignore

About a decade ago I read a book called Against Intellectual Monopoly by Michele Boldrin and David K. Levine – published in 2008, well before anyone typed a prompt into ChatGPT. In the book, Boldrin & Levine used clear data and history to show that, while patents and copyrights were supposedly put in place to fuel progress, they often do the opposite: they hand out monopoly power, raise prices, and slow the next wave of ideas. What held true in the steam era, the VHS boom, and the early streaming days is even clearer now with GenAI.

Intellectual property law still treats knowledge like land or real estate, and as Boldrin & Levine warned, that approach invites the same old rent seeking.

Large AI players scrape first and ask permission later. That tactic is not new. Record labels once did the same with blues masters; software giants did it with user interfaces. The scale is new, though, and the lawsuits are larger. Authors, news groups, and photographers have taken OpenAI, Meta, Stability AI and others to court.

Some things were already wrong, even before AI-generated content…

• Firms spending fortunes on lawsuits and lock‑in strategies that deliver little social value while blocking fresh competitors.
• Pharmaceuticals, software, and even farming tools remaining fenced off decades after first sale, though many patents see no real use.
• Artists facing content removal, account bans, paperwork and legal fees each time they reuse a riff, a verse, or a few seconds of video.

Creative reuse has long been a legal gray zone. Hip‑hop’s rise was fueled by snippets of James Brown, Funkadelic, and many others. After the Bridgeport ruling, US courts said any unlicensed sample, no matter how short, is an infringement. The result was fewer samples, higher costs, and more power for rights holders who could say no. If we apply Bridgeport logic to AI without change, only the biggest labs will afford the clearances. That locks out startups and open research.

Major AI firms aren’t helping the push for IP reform. They treat their datasets as trade secrets, claiming disclosure would ruin their competitive edge. Their resistance to transparency, fair royalty payments, and shared datasets highlights the flip side of the problem. When these firms advocate for IP law “deletion” or lobby for special privileges, they engage in rent-seeking behavior, much like patent trolls.

Disclaimer: I’m not a lawyer, just an AI builder and practitioner (if you’re a lawyer or policy-maker reading this, I’d appreciate your insights – use the contact form or DM me on LinkedIn). These are just my personal thoughts, inspired by the book referenced above.

Shouldn’t we have…

1) Revenue pools or profit-sharing for creators?

2) Antitrust policies to prevent data/patent hoarding?

3) Laws to limit patent overreach?

4) Mandatory transparency?

Building on these considerations, a couple of ideas…

Opt-out / opt-in switch: Wouldn’t it be great to grant creators a switch that says: “You may train on my work for research, but you must license it for commercial systems that replace me”? – pairing it with a full opt-out option (training prohibited). This would mirror the safe‑harbor model of web hosting: broad freedom paired with a takedown tool. The registry could be public, updated in real time, and interoperable via an open API.

Transparency: For all the flak the EU AI Act has been getting (and it does seem some errors were made), I believe the EU is showing the world the way forward on AI transparency, by requiring foundation-model providers to publish summaries of copyrighted data in their training sets. While I agree with criticism that it doesn’t define what constitutes a useful summary or how creators can audit it, I think this transparency requirement should be extended globally and strengthened. The rule should be machine-readable (JSON or similar), and honestly I even believe it should evolve from disclosure-only to disclosure-plus-sanctions, imposing statutory damages when firms train on works that are flagged as “full opt-out” (provision for injunctive relief only if damages prove inadequate, preventing abusive takedowns).

Fair compensation: When a model’s outputs are used at scale and substitute for an existing market, a small slice of revenue could flow into a collective rights pool. Where individual licensing is impractical, the creator’s default share could equal the statutory‑pool rate minus any direct license fee already paid, avoiding double payments. Wouldn’t this be economically more viable than one‑off lawsuits?

Software and business-method: These patents are now a net drag on innovation. Why can’t we replace them with cash innovation prizes and R&D tax credits for open-source releases that meet community‑audit standards?

Trim, trim, trim: Let’s terminate dead patents and trim the long tail. Many patents live well past their useful life. Let unused or non‑commercialized patents lapse after five years unless the owner shows active production. This cuts trolling and frees ideas.

More antitrust: Refusal to license model weights, essential datasets, or key APIs could be codified as presumptively anti‑competitive if the firm’s share of foundation‑model capacity or data exceeds X%. No exclusivity may extend beyond five years without competition‑authority review. If market dominance, national security, or public‑health concerns are proven, then public-interest override activates.

More startup competition: Reduced compliance fees for models under #X parameters.

Fair‑use and mining exceptions: Non‑commercial research and debugging should always be considered lawful and independent of opt‑out status, provided outputs are not marketed. Only if AI outputs affect the creator’s core market, profit‑sharing activates.

Bottom line…

IP has long needed reform, a need long ignored. Now, AI is making it undeniable. A system better aligned with modern technology should prioritize transparency, fair creator compensation, streamlined patents, and stronger competition policies. It’s time to modernize IP for the AI era.