On May 28, 2026, a German court quietly made one of the most consequential rulings in AI law. The Landgericht München (Munich Regional Court) decided that Google’s AI Overviews are Google’s own content — not third-party search results. This seemingly technical distinction has massive implications for every company building AI products. Let’s break down what happened, what it means, and what you should do about it.

What Happened

Case: Landgericht München, Case 26 O 869/26, May 28, 2026.

Google’s AI Overview feature — those AI-generated summaries that appear at the top of search results — falsely linked publishers to scam operations. The affected publishers sued.

Google’s defense was predictable: “We’re just a search engine aggregating third-party content. We’re not the author. Users can fact-check by clicking through to sources.”

The court rejected every part of this defense.

The Three Key Rulings

1. AI Overviews Are Google’s Own Content

The court ruled that AI-generated summaries are Google’s original content, not a reorganization of third-party search results. This is the big one.

Why it matters: Search engines have historically enjoyed special liability protections. They’re treated as intermediaries — they don’t create content, they just index and display it. Google has relied on this distinction for decades.

But AI Overviews are different. The AI synthesizes, paraphrases, and generates new text. The information in the AI Overview didn’t appear in any of the linked sources — it was fabricated by the AI. This isn’t aggregation. It’s generation.

The court said: if you’re generating content, you’re the author. Author liability applies.

2. “Users Can Fact-Check” Defense Rejected

Google argued that users could verify information by clicking through to the actual sources. The court said no — you can’t generate potentially defamatory content and then shift the burden to users to figure out it’s wrong.

This matters because every AI provider has some version of this defense: “We tell users to verify,” “We add disclaimers,” “We link to sources.” The Munich court says that’s not enough.

3. AI Opinions Get Less Free Speech Protection

Perhaps the most philosophically interesting ruling: the court stated that AI-generated opinions deserve less free speech protection than human opinions because they are “the result of an algorithm, not an acquired conviction.”

Free speech protections exist because humans form opinions through experience, reasoning, and belief. An AI doesn’t have convictions — it has statistical correlations. The court decided this distinction matters legally.

Who’s Affected

This isn’t just about Google. The ruling’s logic applies to any AI system that generates content presented to users as information:

• Google AI Overviews — directly affected
• ChatGPT/OpenAI — generates content presented as authoritative
• Claude/Anthropic — same
• Perplexity — literally generates AI summaries from sources (very similar to AI Overviews)
• Microsoft Copilot/Bing Chat — AI-generated answers in search
• Any product with AI-generated text — potentially in scope

If your product generates text that makes factual claims about real people or companies, this ruling says you’re the author of those claims and liable for their accuracy.

For developers building AI features into products, this changes the risk calculus significantly. See our guide on AI liability for developers for broader context.

The DSA Host Provider Defense Is Dead (for AI)

One important technical detail: Google also tried to claim protection under the EU Digital Services Act (DSA) as a “host provider.” Host providers have limited liability for user-generated content they host.

The court ruled that DSA host provider protections don’t apply to AI-generated content because the content isn’t generated by a third party — it’s generated by Google itself. You can’t be both the author and the neutral host of the same content.

This closes what might have been the most attractive legal shield for AI companies in the EU.

Connection to the EU AI Act and Product Liability Directive

This ruling doesn’t exist in isolation. It’s part of a rapidly evolving European legal landscape:

EU AI Act (effective August 2026): The EU AI Act creates a comprehensive regulatory framework for AI systems. High-risk AI systems face strict requirements around transparency, accuracy, and human oversight. While AI Overviews may not be classified as “high-risk,” the Act’s transparency requirements apply broadly.

EU Product Liability Directive (transposition deadline: December 2026): This is the big hammer. The revised Product Liability Directive explicitly includes AI systems as products. This means strict liability — if your AI product causes damage, you’re liable regardless of whether you were negligent.

Strict liability is a game-changer. Under current law, plaintiffs must prove the AI company was negligent (knew or should have known about the error). Under the Product Liability Directive, they just need to prove: (1) the product was defective, (2) they suffered damage, and (3) the defect caused the damage.

No need to prove the company was careless. The AI said something wrong and it hurt someone? That’s enough.

For the full timeline on EU compliance deadlines, see our EU AI Act August 2026 deadline guide.

International Reach

Germany is an EU member state. EU law influences all 27 member states. The Munich court’s reasoning can be cited in other German courts and influence courts across Europe.

But beyond Europe:

• UK: Post-Brexit but still harmonized with EU product liability concepts
• US: While US law is different, persuasive reasoning from foreign courts can influence common law development
• Any jurisdiction with product liability: The “AI as product” framing travels

Companies that operate globally (which is all major AI providers) can’t just ignore one jurisdiction’s rulings. They’ll need to design for the most restrictive regulatory environment they operate in.

What This Means for Developers

If you’re building products with AI-generated content, here’s what you should be thinking about:

1. You Might Be Liable for Your AI’s Outputs

If your product generates text that makes factual claims, and those claims are wrong and harm someone, this ruling suggests you’re liable as the author of those claims. “The AI made it up” is not a defense — you deployed the AI.

2. Disclaimers Probably Aren’t Enough

The “users can fact-check” defense was explicitly rejected. Slapping a “AI-generated, may contain errors” disclaimer on your outputs might not shield you from liability for defamatory or harmful content.

3. Content About Real People/Companies Is Highest Risk

The Munich case involved false claims about real publishers. AI hallucinations about real entities carry the highest legal risk. If your AI says “Company X is involved in fraud” and that’s fabricated, you’ve potentially defamed Company X.

4. Consider Content Filtering and Verification

Practical steps:

• Implement entity recognition to flag claims about real people/companies
• Add verification layers for factual claims
• Consider restricting generation about specific topics (legal claims, health claims)
• Log and monitor outputs for potential harmful content

5. Track the EU Product Liability Directive

By December 2026, EU member states must transpose the Product Liability Directive. This will establish strict liability for AI products across all of Europe. If you’re shipping AI products to EU users, you need to prepare.

For broader legal compliance guidance, see our guides on open source AI legal compliance and who owns AI-generated code.

The Open Source Question

Does this affect open-source model providers differently? It’s nuanced:

• Model providers (Cohere, Meta, Mistral): Release weights, don’t serve end users directly. Likely not liable for downstream use.
• API providers (OpenAI, Anthropic, DeepSeek): Serve content directly to users. Likely liable under this ruling’s logic.
• Developers using models in products: You’re the one presenting AI content to users. You’re likely the liable party.

If you use an open-source model in your product, you can’t shift liability to the model creator. You chose to deploy it, you’re the author of its outputs in the context of your product.

The AI copyright and training data landscape is equally complex and worth understanding alongside this ruling.

What Should You Do Right Now?

Immediate actions:

1. Audit your AI-generated content — Where does your product present AI text as factual information?
2. Identify highest-risk outputs — Claims about real people, companies, or products
3. Implement output monitoring — Log AI outputs, flag potential defamatory content
4. Review your terms of service — Ensure they reflect the reality that you’re generating content, not just hosting it
5. Talk to legal counsel — This ruling is new and its full implications are still developing

Medium-term actions:

6. Build content safety layers — Entity-aware output filtering
7. Prepare for Product Liability Directive — December 2026 transposition deadline
8. Consider EU-specific deployment strategies — Different risk profiles for different markets
9. Document your AI safety measures — Demonstrates due diligence if challenged

The Bigger Picture

This ruling is one of the first concrete legal decisions treating AI-generated content as the AI provider’s own speech. It won’t be the last.

The trend is clear: courts and regulators are moving toward holding AI companies liable for their outputs, just like any other publisher or product manufacturer. The era of “it’s just an algorithm, we’re not responsible” is ending.

For developers, this means AI safety isn’t just an ethics question anymore — it’s a legal one. The cost of harmful AI outputs now includes potential court judgments, not just reputational damage.

Stay informed. Build responsibly. And talk to a lawyer before shipping AI features that make claims about real entities.

FAQ

Does this ruling apply outside Germany?

Directly, it only applies within the German legal system. But EU law tends to harmonize across member states, and the EU Product Liability Directive (which explicitly covers AI) must be transposed by all EU countries by December 2026. Other courts will likely follow similar reasoning. AI companies operating in the EU should treat this as a preview of broader liability.

Am I liable if I use an open-source model in my product?

Likely yes. You’re the one deploying the model and presenting its outputs to users. The model creator (e.g., Cohere releasing North Mini Code under Apache 2.0) isn’t directly serving content to your users — you are. You’re the “author” in the court’s framework.

Do AI disclaimers and safety warnings protect me?

The Munich court explicitly rejected Google’s “users can fact-check” defense. While disclaimers might reduce damages or demonstrate good faith, they probably don’t eliminate liability for genuinely harmful content. You can’t defame someone and then say “but we warned users we might be wrong.”

What’s the difference between the EU AI Act and the Product Liability Directive?

The EU AI Act is a regulatory framework that classifies AI systems by risk and imposes requirements (transparency, accuracy, oversight). The Product Liability Directive is about compensation — if an AI product causes damage, victims can sue for strict liability. The AI Act is about prevention and compliance; the Product Liability Directive is about remedy after harm.

Should I stop building AI features for EU users?

No. But you should build them responsibly with appropriate safeguards. The EU market is too large to ignore. What you should do: implement content safety measures, avoid generating unchecked factual claims about real entities, monitor outputs, and prepare for the Product Liability Directive deadline. Building defensively now saves expensive retrofitting later.

How does this affect AI coding assistants specifically?

Coding assistants are lower risk than general-purpose AI for this specific ruling because they generate code, not factual claims about people or companies. However, if your coding assistant generates comments that reference real libraries, companies, or individuals incorrectly, there’s theoretical risk. The bigger concern for AI coding tools is the ownership of AI-generated code question.