The MFE v. Perplexity AI Copyright Case — Grievances, Evidence Quality, Consequences, and Lessons for AI Makers

by ChatGPT-5.1

The lawsuit filed by MFE–MediaForEurope against Perplexity AI marks a watershed moment in European AI-copyright law. It is the first Italian court case targeting AI training on audiovisual works, and it arrives in the context of a rapidly expanding global litigation wave against Perplexity. The OpenTools summary of the dispute and a Bloomberg Law report confirm that the case is now formally before the Civil Court of Rome, led by MFE subsidiaries Reti Televisive Italiane (RTI) and Medusa Film. The litigation alleges large-scale, unauthorized ingestion of copyrighted film and TV works to train Perplexity’s models.

The dispute is not merely about infringement but about the boundary between AI development and media rights in the EU’s newly emerging post–AI Act enforcement landscape. Understanding the grievances, the evidentiary posture, and the potential consequences for Perplexity offers a preview of the regulatory and commercial world AI companies now face.

1. The Grievances: What MFE Says Perplexity Did Wrong

The grievances fall into four main categories, all consistent with allegations already filed against Perplexity in the United States and Japan.

1. Large-Scale Scraping of Audiovisual Content Without Permission

MFE alleges that Perplexity:

• “Systematically scraped and utilized copyrighted audiovisual and film works”— including full movies and TV programming owned by RTI and Medusa — to train its AI models without licenses or permissions.

• Violated Italy’s copyright law, which grants strong exclusive rights over reproduction, extraction, and processing of film content.

Unlike many U.S. cases, which mostly concern text, this complaint involves video and film — assets for which rights-holders maintain exceptionally high commercial and territorial control.

2. Use of Copyrighted Media Beyond Any Permitted Exception

Italy does not have a broad “fair use” doctrine. It operates under strict, closed-list exceptions. Training a commercial AI system on full-length films and shows would not fall into any recognized exception, especially if done at scale.

Thus, if the ingestion occurred, infringement is almost automatic.

3. Lack of Transparency and Lack of Licensing Infrastructure

MFE stresses that Perplexity:

• Never disclosed the use of their materials,

• Never sought a license,

• Never engaged in data governance dialogue despite widespread industry concerns.

This grievance echoes complaints by the BBC, News Corp, Dow Jones, Nikkei, Asahi Shimbun, and Yomiuri Shimbun, all of whom have launched lawsuits between 2024–2025 because Perplexity allegedly ignored licensing overtures and used training corpora derived from “surreptitious or illicit scraping.”

4. Ongoing Harm: Market Substitution and Derivative Value Loss

MFE argues that Perplexity’s use creates:

• Loss of control over exploitation rights,

• Competitive harm, because the model can summarize or reproduce expressive content,

• Erosion of licensing markets, particularly for clips, archive uses, and derivative rights.

The plaintiffs therefore seek recognition of illegality, injunctions, damages, and daily fines for future violations.

2. The Evidence: How Strong Is MFE’s Case?

A. The Evidence We Know They Have

Although the full complaint is not yet public, the OpenTools and Bloomberg reports confirm that the following evidence is central:

• Claims of systematic scraping — implying logs, server traces, or content appearing in model outputs.

• Model behavior demonstrating knowledge of protected film content, e.g., detailed scene summaries not available in public domain sources.

• Similarity in language or structure between Perplexity answers and copyrighted scripts or subtitles.

• Precedent from U.S. litigation where federal courts have already dismissed Perplexity’s attempts to claim search-engine–style safe harbors.

Because audiovisual works are more uniquely identifiable than text, proving machine-learning ingestion is easier: models often reveal memorized sequences or perform highly specific recall tasks when prompted.

If MFE lawyers can show that Perplexity’s model:

1. recites scenes,

2. reproduces subtitles,

3. summarizes non-public clips, or

4. identifies dialogue unique to MFE’s films,

then the evidence of copying becomes hard for any judge to ignore.

B. Likely Strength Under Italian and EU Law

The lawsuit arrives at a crucial regulatory moment:

• The EU AI Act requires GPAI providers to document training data and honor opt-outs for copyrighted materials.

• The Italian judiciary historically interprets copyright expansively.

• Italy’s audiovisual sector has political importance (via Mediaset/MFE and the Berlusconi family), which increases institutional focus.

Taken together, these make the evidentiary burden lighter for MFE than for plaintiffs in the U.S.

C. Perplexity’s Defense: Weak or Strong?

Perplexity has consistently used the “we operate like a search engine” defense.

But:

• U.S. courts have already rejected that framing.

• Unlike Google, Perplexity has no settled case law supporting broad snippet reproduction.

• Perplexity keeps its training data sources opaque, which courts often interpret negatively.

• Unlicensed audiovisual ingestion is harder to defend than text ingestion.

In short: Perplexity’s evidentiary position appears weak, and the plaintiffs’ case is substantially strengthened by the EU’s regulatory environment and the nature of the copyrighted works involved.

3. Consequences for Perplexity: Legal, Financial, and Strategic

This case may become existential for the company. The consequences fall into six categories:

1. Injunctions Blocking Model Distribution in the EU

If MFE wins, Perplexity could face:

• EU-wide injunctions against models trained on infringing data,

• Orders requiring retraining or “purging” of weights,

• Restrictions on deployment in Italy, with ripple effects across the EU.

The EU AI Act amplifies this risk: non-compliant GPAI providers can be fined up to 7% of global turnover.

2. Substantial Damages and Daily Penalties

MFE is asking for:

• Monetary damages,

• Continuing per-day fines for future infringements.

Italy commonly awards injunction-linked penalties (“astreintes”), which can reach millions of euros if the conduct continues.

3. Forced Disclosure of Training Data

Courts may compel:

• Disclosure of training corpora,

• Disclosure of data-pipeline architecture,

• Audit access to logs and model internals.

This is a major threat to Perplexity’s competitive position — and one reason most AI companies settle early.

4. Collapse of the “search engine” positioning narrative

If an Italian court rejects Perplexity’s analogy to Google or Bing, this will undermine:

• Their public marketing,

• Their defense in U.S. litigation,

• Their ability to argue for DMCA-style safe harbors.

5. European publishers will follow

If MFE succeeds, expect lawsuits from:

• RAI,

• Sky Italia,

• RTL Group,

• Vivendi/CANAL+,

• The European Publishers Council members.

A win would shift the bargaining power towards mandatory licensing.

6. Investor and partner flight

Rising litigation costs and regulatory exposure could:

• Raise questions about Perplexity’s viability,

• Drive up insurance premiums,

• Create acquisition risk or force restructuring.

Perplexity is already facing simultaneous lawsuits from News Corp, Dow Jones, BBC, Nikkei, Asahi, and Yomiuri — meaning MFE’s case compounds an already precarious situation.

4. How AI Makers Could Prevent Cases Like This

The Perplexity case is a symptom of a deeper industry problem: training models in a pre-licensing era and hoping the law would eventually validate the practice. That gamble is failing globally.

To prevent similar disputes, AI companies must shift from the “scrape first, defend later” paradigm to a “license, document, and comply” paradigm.

1. Adopt Licensing-First Data Acquisition

AI firms should:

• License film, TV, music, books, and journalism content directly from rights-holders.

• Participate in emerging collective licensing regimes (e.g., under EU AI Act Article 53 mechanisms).

• Embrace institutional data marketplaces.

2. Maintain Transparent Training Data Documentation

Under the EU AI Act, transparency is mandatory.

Companies should:

• Publish high-level training-data categories,

• Maintain internal audit trails,

• Provide documentation to rights-holders when requested.

3. Build and Respect Machine-Readable Opt-Out Systems

The EU is developing protocols for rights-reservation under TDM exceptions. AI companies must support:

• robots.txt (and successors),

• schema.org rights metadata,

• IETF-approved opt-out protocols.

4. Implement Content Filtering Before Training

State-of-the-art preventive measures include:

• Perceptual hashing to detect copyrighted clips,

• Watermark recognition,

• Rights-owner registry matching,

• Automated license-checking pipelines.

5. Shift to Licensed Synthetic Data for Sensitive Domains

Instead of using unlicensed film archives, AI developers can:

• Generate synthetic audiovisual corpora based on licensed foundation datasets,

• Use simulation-based training for model robustness.

6. Develop Clean-Room Training Processes

Under a clean-room approach:

• Scraped data is analyzed only to identify categories but not stored,

• Only licensed material enters persistent training sets,

• Model weights are attestable as non-infringing.

7. Build Active Relationships With Media and Publishing Sectors

AI firms must stop treating rights-holders as adversaries.

Regular engagement with media groups, publishers, and collecting societies reduces litigation risk and encourages negotiated solutions.

Conclusion

The MFE v. Perplexity AI lawsuit is likely to become a precedent-setting case in Europe, particularly because it concerns audiovisual content — a domain with high commercial value and strong legal protections. The grievances are clear: large-scale unlicensed use of films and TV content, lack of transparency, disregard for opt-out mechanisms, and economic harm. The evidence appears strong, especially because courts in the U.S. are already rejecting Perplexity’s defenses and because audiovisual recall is easier to prove than text ingestion.

The consequences for Perplexity could be severe: injunctions, damages, forced retraining, regulatory sanctions under the EU AI Act, and a broader collapse of its “search engine” narrative. For the entire AI industry, the case reinforces that the era of unlicensed scraping is ending. The path forward requires licensing, transparency, robust data governance, and collaboration with rights-holders.

If AI firms fail to adapt, this case — Italy’s first — will be only the beginning of a long wave of global enforcement.