gafam.ai
Five Companies.
One Planet.
Artificial Intelligence.

How the company engages with European rules and users — fines, investigations, compliance, and constructive moves. Higher is more respectful.

How the company treats users worldwide — business model, privacy, lock-in, openness, and documented harms. Higher is more respectful. An editorial estimate.

Google carries by far the heaviest European enforcement record of the five — a record €890 million DMA fine in 2026, a €2.95 billion adtech penalty, the €4.1 billion Android case, and an open antitrust probe into its AI training — and compliance has come mostly under pressure, not by choice. Its users get genuinely useful free services, but the underlying model runs on data and advertising, and missteps like its withdrawn Earth deepfake tool show speed outrunning care. On the future, DeepMind runs arguably the strongest safety-research operation of the five, with a published Frontier Safety Framework — yet the product side keeps outrunning the research, from that one-day deepfake tool to a phone assistant that now reads every app on your screen.

Apple was fined €500 million under the DMA for anti-steering and has resisted interoperability orders more stubbornly than any peer, repeatedly delaying AI features in the European market. Its users benefit from the strongest privacy reputation of the five — a real advantage — but one paid for through lock-in, poor repairability, high prices, and a tightly closed ecosystem. On the future, caution is its whole strategy: slow deployment, on-device processing, and a considered privacy architecture — though this is safety-by-slowness more than frontier-safety leadership, since Apple builds few frontier models of its own and leans on partners instead.

Meta drew a €200 million DMA fine and a long fight over its "pay or consent" model, but has since agreed a revised advertising scheme with Brussels, and its open-weight models are a genuine contribution to European AI sovereignty. For its users, the data-and-advertising engine and well-documented social harms weigh heaviest of the five; its open-source generosity lifts the score only slightly. On the future it lands lowest, and it is the hardest case: open weights are genuinely double-edged — transparent and empowering, but a permanent loss of control over misuse — its own model breached an outside company during testing this summer, and it frames open weights as a competitive weapon against regulation. Open disclosure of risk levels counts in its favour, but not enough to outweigh the rest.

Amazon has avoided major EU fines so far, though AWS is under a DMA market investigation over whether it should carry full gatekeeper obligations. Its users get unmatched convenience, offset by labour-practice concerns, marketplace self-preferencing, and extensive data collection across its retail and device ecosystem. On the future, it sits in the middle largely by having less frontier exposure — more an AI infrastructure and marketplace provider than a frontier lab — and its large investment in a safety-focused model developer points in a constructive direction, though it shows little safety leadership of its own and has yet to define the responsibilities of being a main channel through which models reach the world.