AI Training Data Fines in Europe Top 13 Billion Zloty as AI Act Deadline Nears

Anthropic, Meta, Google, Clearview AI, Apple, Amazon and OpenAI have together paid more than $3.5 billion, roughly 13 billion zloty, in fines since 2022 for using personal data to train artificial intelligence systems without proper consent or legal basis. The data was compiled by analytics firm Surfshark, first reported in Poland by Rzeczpospolita.

Surfshark's tally covers ten regulatory and court proceedings in Europe involving Anthropic, Meta, Google, Clearview AI, Apple, Amazon and OpenAI. The methodology included only cases with a clearly identified company, a quantifiable financial penalty, and a direct link to AI practices, such as illegal use of training data or misleading marketing claims.

Highest Fines

Topping the list is Anthropic, fined $1.5 billion in 2025 for training language models on pirated copies of books. Close behind is Meta, fined $1.4 billion in 2024 for collecting biometric data for facial recognition systems without user consent. Google paid $291 million in 2024 for training AI on copyrighted media content without authorization, while Apple was fined $250 million in 2026 for overstating its artificial intelligence capabilities.

Clearview AI racked up four separate fines totaling more than $100 million from data protection authorities in the Netherlands, Italy, France and Greece for unauthorized scraping of facial images from the internet. The company is the only one refusing to pay, arguing that European authorities have no jurisdiction over it. Amazon paid $25 million in 2023 for storing voice recordings of children using Alexa without proper consent.

Not All Fines Are Final

The tally also shows that some fines have not survived judicial review. A $17 million fine imposed on OpenAI in 2024 for training ChatGPT without a legal basis was overturned by an Italian court in 2026. That's a sign regulators are still testing the limits of enforcement, and that tech companies are successfully appealing some rulings.

This may just be the beginning. The overall trend shows that accountability is starting to catch up with innovation, and the industry needs to rethink both how it builds AI and how it sells it. - Dr. Luis Costa, research lead at Surfshark

August Threshold for the AI Act

Fines to date have mainly been imposed under the EU's General Data Protection Regulation (GDPR) and national copyright law, not the AI Act itself. That's set to change on August 2, 2026, when the full penalty system under the EU's Artificial Intelligence Act takes effect, covering high-risk systems and general-purpose models. The maximum fine will reach 35 million euros or 7 percent of a company's global annual revenue, whichever is higher.

According to figures cited by Rzeczpospolita, only 36 percent of European companies say they are well acquainted with the new rules, suggesting implementation is still at an early stage. At the same time, Brussels is working on the Omnibus AI package, intended to simplify parts of the AI Act, underscoring the tension between data protection and keeping European companies competitive with the US and China.

What It Means for Poland

For Polish companies deploying AI tools, this means double regulatory risk: existing personal data protection rules and the coming, far stricter penalty regime under the AI Act. Surfshark's tally shows that even the biggest market players, with extensive legal departments, made mistakes in sourcing training data. That suggests smaller companies should already be auditing the provenance of data used to train or fine-tune their own models, before the August deadline makes violations significantly more costly.

Sources: Big Tech slapped with $3.5bn in fines for using your personal data to train AI (techradar.com), Tech companies hit with $3.5B in AI fines since 2022, led by Anthropic and Meta (digitalinformationworld.com), Gigantic fines for Big Tech. The EU is slowing down AI development (rp.pl)