IP as bargaining power in the age of AI: the only model that matters

A well-known financial journalist at a major business publication told me something that should terrify (and wake up) anyone who still thinks paywalls are a strong moat. He regularly publishes a deep-dive feature behind a paywall. Within minutes, every time, the same reporting (his words) appears on an AI-generated website: no permission, no attribution, no payment.

No bueno.

Another prominent journalist/author described a different version of the same problem: with a few prompts, an LLM can reproduce large portions of his books. Work he never licensed for model training, never authorised for reuse, and never got paid to “contribute” to.

Both stories point to the same commercial reality: your work can now be turned into a competing product faster than you can file an invoice. And because the copying can happen through a chain (scrape → train → summarise → rehost) the theft feels diffuse, almost deniable.

So what to do?

Let’s strip this down to what actually matters for independent publishers and freelance journalists.

The question is less whether AI is “good” or “bad”. Rather, the question is whether you have any bargaining power when your reporting and archives are valuable to someone else’s model and someone else’s product.

In the current media landscape, we have completed the transition from an economy of information scarcity to one of acute attention scarcity. As Herbert Simon famously posited, a wealth of information creates a "poverty of attention."

For long-form creators, those producing the deep-dive reporting, B2B analysis, and investigative archives that require high fixed costs, generative AI acts as a dual-force multiplier. It renders "good enough" substitutes infinitely cheap to produce, yet simultaneously elevates the premium on rights-cleared, high-signal content.

In this environment, legal rights and provenance become the only "enforceable scarcity" remaining, and are no longer mere administrative formalities.

High-quality, or “high-fidelity”, content acts as a trusted signal in a digital ecosystem increasingly cluttered with AI-generated noise. The strategic imperative for media executives and independent creators is to recognise that Intellectual Property (IP) provides the formal frame around these attention-based assets, transforming them from vulnerable public goods into excludable, high-value leverage.

This week we look at a simple way to think about this problem: three actors, two types of value, and two legal environments.

First, the players in the arena:
Writers/Creators/Publishers, AI developers, and Audiences/Customers.

AI developers can extract value from your work in two ways:

OK, now we know who the stakeholders are, and what’s happening. Next, we need to acknowledge the uncomfortable reality of the situation: whether training is legal or otherwise, and how outputs/responses still hurt demand for the work that made the system useful in the first place.

There are two environments that determine whether your “no” has any teeth.

Environment 1: training is permitted by default (“free-ish”)
Here, training is mostly treated as fair use or allowed text-and-data mining (TDM) because you didn’t reserve rights. For example: permissive fair use in the US, and the EU’s commercial TDM exception where no opt-out is used.

In this world, if you refuse to license, the AI firm may still train anyway, so your refusal has little effect.

Environment 2: training requires permission (“permissions/licensing”)
Conversely, here training on your content generally needs your express permission (authorisation), because the jurisdiction is stricter, or because you reserved rights/opted out, or because database rights apply.

Examples include the UK under current law and EU cases where rights are properly reserved or database rights are invoked. In this world, if you refuse to license, the AI firm must either train without your material (and accept lower performance/value) or pay to access it.

This distinction explains, to some degree, why publishers and AI labs are talking past each other. AI labs argue inside a world where permission is optional; publishers argue inside a world where permission is the price of admission.

To understand the capture of value, we must define the variables within the negotiation between a Creator and an AI Developer.

Let’s use plain Nash-style bargaining logic. Translation: cooperative bargaining; a deal happens when there’s a surplus to split.

Here’s the hard part: your share of that pie depends less on how loudly you complain and more on four conditions you are able to lever.

The comparative statics are brutal: IP is the exclusive mechanism for capturing the V surplus that would otherwise default to AI labs.

In the “free-ish” training world, the your bargaining power as a creator is significantly diminished. Without an enforceable mechanism, the marginal cost of reproduction for your work becomes zero for the developer, effectively stripping you of the ability to charge above that cost, and they simply use the data without paying

In the permissioned world, your bargaining power rises only if your corpus is hard to replace, AI firms can’t route around you, and creators can coordinate.

That’s the honest, if brutal, core of this “IP as bargaining power” idea. It’s less a guarantee of payment, and more the rule-set that determines whether payment is even a rational outcome.

Even if training stays “free-ish,” demand for your work is still impacted in three ways: substitutes, complements, or neutral effects.

This matters because rights are most contested at the output layer when the copying comes to the surface: i.e. rights can be used to challenge verbatim or near-verbatim reproduction of paywalled or valuable expression, and to monetise high-fidelity summaries or syndication, but not to block competition of ideas more generally.

In other words: you can’t stop an AI product from answering “What is GDPR?” But, you can fight it when it starts laundering your paywalled reporting into a substitute product, and earning revenue from it.

Some arguments treat AI-driven “market dilution” (i.e. more “good enough” supply lowers the value of originals) as a cognisable harm; others view that as an overreach that risks turning ordinary competition into infringement. I understand this to be disputed territory and my advice is not to build strategy on a single theory. The safer ground is observable substitution, enforceable rights reservations, and contract terms that define lawful access and reuse.

Copyright is merely the foundational layer of a more sophisticated defensive architecture. For long-form publishers, leverage is derived from several hidden levers:

• Contractual Controls (T&Cs)

• Trademarks and Brand Equity

• The Database Right (Directive 96/9/EC)

These levers provide protection if AI training is deemed fair use, shifting the focus from the act of training to the act of access. This multi-layered approach transitions the conceptual value of a brand into a hard reality of global economic leverage.

If you’re looking for the takeaway to hold onto: rights don’t print money. They determine whether your “no” has any consequences. And in this market, consequences come from the three things we keep returning to: distinct archives, coordination, and the ability to prove and enforce the terms of access across borders.

Part 2 next week maps the US/UK/EU triangle: the point where the same underlying idea produces three different fights.