France Fined Google $270 Million for Training Gemini on News It Never Licensed
What happened
A EUR 250 million fine for a training dataset reads like an unusual penalty in tech, where content has historically moved across platforms with minimal friction. But the decision France's Autorité de la Concurrence issued in March 2024 was specific: Google had used articles from press agencies and media companies to train its AI chatbot, Bard (since renamed Gemini), without obtaining permission from the content's owners first. The USD 270 million equivalent was not an abstract regulatory gesture. It was a statement that the inputs to a commercial AI system are subject to the same authorization requirements as any other licensed use of protected content.
The fine emerged from a competition investigation rather than a copyright lawsuit. French law had established that publishers hold enforceable rights over how their content is used commercially, including by platforms that aggregate or index it. Google had previously entered into commitments with the regulator about how it would handle those rights in the context of its services. When the authority found that Google had extended its use of publisher content into AI model training without first securing new authorization, those prior commitments made the conduct difficult to contest. Google reached a settlement and described the outcome as resolving a dispute that had lasted too long, which was one way to characterize a fine that measured the gap between what was permitted and what was done.
The EUR 250 million figure put a number on something the industry had long treated as settled: that training on crawled web content is legally neutral. News publishers had spent years arguing that their articles were not simply data available for any commercial purpose. When that argument was applied to AI training pipelines, it landed with enough force to extract a nine-figure penalty from one of the largest technology companies in the world. The French case was among the more visible enforcement actions of this kind, but it was not the only one developing across European jurisdictions.
What the case also revealed was the process that did not exist before the training runs took place. There was no framework at Google, at least none on the record, for clearing publisher content against a rights ledger before feeding it into a model. The default assumption in AI development has been that crawled data is trainable data unless a rights holder erects a specific technical barrier. France's competition authority said that assumption was wrong when it came to press content, and the fine confirmed the point in the clearest terms available.
The deeper gap the case exposes is one of documentation rather than law. Even after the settlement, there is no public register of what data trained Gemini, under what authorization, or when those decisions were made. The French fine covered content governed by French law, representing one slice of what the system absorbed. A provable record of what a model trained on, together with documentation of the legal basis for each inclusion, would have made that scope auditable from the start instead of a matter of regulator investigation, litigation, and delayed settlement. Without that record, the full extent of unlicensed training remains a question with no reliable answer.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.