Google can train AI on content even if publishers opt out: exec

Google can train its search-specific AI products, like AI Overviews, on content across the web even when publishers have chosen to opt out of training its AI products. Google DeepMind vice-president Eli Collins testified in court on May 2.

This admission came during an ongoing antitrust trial, where the US Department of Justice (DOJ) seeks to curb Google’s dominance in online search.

The DOJ has asked a federal court to order Google to divest its Chrome browser and stop paying to be the default search engine—including on AI tools like Gemini.

DOJ attorney Cristina Aguilar presented evidence showing that Google’s search data could significantly boost its AI models.

An internal document from August 2024 showed that Google filtered out 80 billion of 160 billion content tokens after removing data from publishers who opted out of AI training.

Google’s lawyers argued that search dominance doesn’t restrict AI competition, noting that chatbot developers often make direct deals with content providers.

🔗 Source: Bloomberg

🧠 Food for thought

1️⃣ The DOJ’s proposed Chrome divestiture echoes historic tech breakups

The DOJ’s recommendation to force Google to sell Chrome draws parallels to previous landmark antitrust cases that restructured American industries.

This remedy mirrors the 1982 AT&T breakup, when the telecom giant was forced to divest its regional “Baby Bell” companies after an 8-year legal battle, fundamentally reshaping the telecommunications landscape 1.

Similarly, Microsoft’s antitrust case in the late 1990s centered on browser bundling—specifically how it tied Internet Explorer to Windows to crush Netscape—though Microsoft ultimately avoided a breakup through a settlement that imposed business restrictions 2.

These past cases demonstrate that structural remedies aren’t unprecedented in antitrust enforcement, though they’re typically reserved for the most severe monopoly situations.

The current focus on Chrome is strategic because browsers serve as crucial gateways to the internet, with Google paying approximately $20 billion annually to be the default search engine across various platforms, a practice the DOJ now seeks to restrict.

2️⃣ The battle over AI training data exposes publisher power dynamics

The revelation that publisher opt-outs removed 50% of Google’s potential AI training data (80 billion of 160 billion tokens) demonstrates the significant leverage content creators now wield in the AI development ecosystem.