If the EU’s top court embraces an accessibility standard for “communication to the public,” copyright enforcement becomes less dependent on proving that a service “targeted” a country...

How Anne Frank will (also) have a say on copyright enforcement in the age of AI

by ChatGPT-5.2

This piece by Eleonora Rosati on The IPKat discusses a question at the heart of online copyright enforcement in Europe: when something is put online, which country’s copyright law applies—and how hard should it be for a rightsholder to sue?

It focuses on an Opinion by Advocate General Rantos in a case involving Anne Frank (Dutch referral; C-788/24). The key issue is whether making a work available on a website counts as “communication to the public” in an EU Member State even when the upload wasn’t aimed (“targeted”) at that country—for example, where the site is not in that language, doesn’t market there, and maybe the uploader says “I didn’t mean EU audiences.”

The punchline

AG Rantos’ view (as summarized in the post) is essentially:

• If the content is accessible in a Member State, it can be treated as “communication to the public” in that Member State.

• In other words, you don’t have to prove targeting to satisfy the “communication to the public” element.

The post then explores why that matters, and argues it could make enforcement easier—especially in an AI context.

Accessibility vs targeting: the difference that changes everything

The essay contrasts two ways to “locate” an online act:

1) Targeting approach (harder for rightsholders)

A court asks: Was the activity meaningfully aimed at this country?
Evidence might include language, marketing, delivery arrangements, pricing in local currency, customer support, etc. (The post references this kind of analysis via case law examples.)

This approach is attractive to online services because it reduces “global liability by default.” If you didn’t aim at France, you’re less likely to be sued under French law.

2) Accessibility approach (easier for rightsholders)

A court asks: Can people in this country access it?
If yes, then (at least for some legal questions) that country becomes relevant.

The AG’s position, as explained in the post, leans toward accessibility for the “communication to the public” question—meaning online availability can trigger local copyright relevance without a targeting showing.

Do I agree with the view in the post?

Where I agree

I agree with the post’s core intuition: in a world of VPNs, mirrors, and globally reachable platforms, a strict targeting requirement can become a gift to sophisticated infringers. If “not targeting” becomes the easy escape hatch, then the people who suffer are typically:

• authors and publishers facing pirate distribution,

• creators whose works are re-uploaded across borders,

• rightsholders dealing with fast-moving “copy-and-spread” ecosystems.

In that sense, accessibility can restore some realism: if your activity predictably reaches (and harms) the EU public, you shouldn’t automatically get to disclaim EU consequences by pointing to where you first uploaded or which audience you intended.

I also agree with the post’s point that, if the CJEU goes this way, it could interlock with the EU AI Act’s “if it’s made available in the EU, EU copyright compliance matters” logic—especially for general-purpose AI models offered to EU users.

Where I’m more cautious than the post

I’m more cautious on three fronts:

1. Accessibility can turn into “liability everywhere.”
   If mere reachability is enough, almost any online act becomes legally “present” in every Member State at once. That can be fair when you’re dealing with mass infringement at scale—but it can also be blunt and over-inclusive for edge cases, small actors, and borderline factual scenarios.

2. Forum shopping risk is real.
   The post notes a practical truth: even if exclusive rights are “harmonized,” remedies and court practice are not. If accessibility lets rightsholders choose among many venues, you’ll see more strategic filings—sometimes for good reasons (speed, expertise), sometimes to maximize pressure.

3. The split between “applicable law” and “jurisdiction/damages” can stay messy.
   The post itself flags that even where accessibility helps establish a forum, damage recovery may be territorially limited in some configurations (sue where harm is felt → damages limited to that territory). So accessibility might be a helpful lever, but not a complete solution.

Net: I broadly agree with the post’s direction for enforcement practicality, but I think courts will need to craft limiting principles so accessibility doesn’t become an indiscriminate “global strict liability” rule.

Consequences if the CJEU follows this approach

A. Consequences for rights owners (authors, publishers, collecting societies)

1) Enforcement gets easier at the first gate.
Rightsholders may no longer have to assemble an elaborate “targeting” dossier just to establish that an online act counts as communication to the public in that Member State. That matters because speed is everything online.

2) Stronger ability to pick a workable court.
If accessibility expands options, rightsholders can choose courts with:

• faster injunction practice,

• more experience with cross-border IP,

• stronger interim relief,

• better evidence procedures.

3) But also: more procedural chess.
Defendants will respond with their own tactics:

• arguing limited territorial damages,

• challenging connection to the forum,

• shifting disputes into jurisdictional trench warfare.

4) Strategic pressure increases—so settlement dynamics change.
If a defendant faces credible risk of injunctions or quick enforcement steps in multiple Member States, the leverage shifts. That can benefit legitimate rightsholders—but it can also inflate nuisance claims unless courts filter properly.

B. Consequences for scholarly publishers specifically

Scholarly publishing is a special case because the “infringement” question often comes bundled with institutional legitimacy, research integrity, and platform dependence.

1) More leverage against AI systems offered in the EU.
If an AI model is made available to EU users, the post suggests this approach could support the idea that EU copyright compliance is relevant “regardless of where training happened.” That matters because model training, hosting, and corporate domicile are often outside the EU.

2) The fight shifts from “where did training happen?” to “what did the system do?”
The post is clear that rightsholders would still need to prove key facts—especially derivation/regurgitation when alleging outputs reproduce protected expression. So the practical bottleneck becomes:

• evidence of copying or recognizable reproduction,

• attribution of the act (developer vs user vs both),

• mapping which restricted acts occurred (reproduction, communication, etc.).

  The IPKat

3) It increases the value of monitoring and “proof infrastructure.”
For publishers, the biggest operational implication is: you can’t enforce what you can’t prove.
If enforcement becomes more viable procedurally, the differentiator becomes the quality of:

• logging and capture of outputs,

• reproducibility of prompts and results,

• expert analysis showing substantial similarity / recognizable extraction,

• chain-of-custody discipline.

4) It nudges AI vendors toward EU-facing risk controls.
Expect more:

• geofencing and differentiated EU product versions,

• stricter output filtering,

• dataset governance and licensing clean rooms,

• “copyright compliance” paperwork—some meaningful, some theater.

The AI angle: why this is a big deal beyond one diary case

The post’s most important move is connecting the doctrine to AI reality:

• Training is often cross-border.

• Outputs are served instantly into many jurisdictions.

• Users can prompt models to reproduce text “in the style of” or “summarize with quotes,” and in some cases extract near-verbatim passages.

If accessibility is the rule, then EU availability becomes the anchor—not the location of the training cluster. That reduces the incentive for “jurisdiction shopping by infrastructure” (train elsewhere, host elsewhere, claim immunity).

But: the hard problems don’t go away. They concentrate around evidence and attribution:

• When is an output an infringing reproduction vs a lawful summary/idea-level restatement?

• Who “did” the act: the user, the model provider, or both?

• What if a model sometimes regurgitates due to memorization—does that imply unlawful copying in training, unlawful making available at inference, or both?

The post rightly signals that these questions will shape the next stage of enforcement.

My practical takeaways for rights owners and scholarly publishers

1) Treat this as a “procedural leverage” shift, not a silver bullet

If the CJEU adopts accessibility here, it mainly changes the front-end friction of enforcement. The back-end still depends on proof, remedies, and how courts handle cross-border limits.

2) Invest in evidence generation, not just legal theory

For scholarly publishers, the most defensible posture is:

• repeatable testing protocols,

• forensic capture of outputs,

• similarity analysis capability,

• escalation playbooks (notice → negotiation → litigation).

3) Use the moment to reset AI licensing conversations

If EU availability triggers EU copyright expectations more forcefully, publishers can press for:

• clear licensing scope,

• audit/assurance rights,

• provenance and transparency commitments,

• output-guardrail obligations,

• meaningful redress mechanisms for leakage/regurgitation.

4) Expect (and plan for) vendor segmentation

Some AI providers will treat the EU as a compliance zone and ship a “safer” product there. Publishers should anticipate:

• different behavior EU vs non-EU,

• EU-specific terms of service,

• more formalized copyright compliance claims—verify them.

Bottom line

The post argues that if the EU’s top court embraces an accessibility standard for “communication to the public,” copyright enforcement becomes less dependent on proving that a service “targeted” a country—and that this could be especially consequential for AI models made available in the EU. I largely agree with that direction because it matches how online harm actually propagates.

But I’d treat it as a sharpened tool, not an automatic win: the policy upside (less evasive infringement) comes with governance risks (forum shopping, overbreadth, legal uncertainty). For scholarly publishers and other rightsholders, the winning strategy in that world is less about rhetoric and more about proof infrastructure + disciplined enforcement + smarter licensing leverage.