AI Press Rights: Gemini Ruling Plus 4 Updates
EU adviser says chatbots may breach press rights; NYT sues Brave over scraped content.
The Legal Line Moves From Training to Output
Why the NYT vs. Brave Suit Matters for API Builders
The Rest of the Briefing: Power, Bounties, and Work Hours
This week's AI briefing is less about model releases and more about who owns the words those models output. The first generative AI case to reach the EU's top court has an adviser saying chatbots like Gemini can infringe press rights when they reproduce protected articles—and the New York Times has sued Brave over selling scraped journalism to AI companies. Together, these stories signal a legal shift from what AI learns to what it repeats.
## The Legal Line Moves From Training to Output
Advocate General Maciej Szpunar's non-binding opinion in Case C-250/25 says a chatbot that retrieves and reuses a news article to answer a user may violate reproduction and communication-to-public rights, especially if it bypasses a paywall or anti-scraping measures, or if it reuses wording so recognizably that the reader no longer needs the original. Facts remain unprotected, and using protected material to train a model is generally allowed under EU text and data mining exceptions. For developers, the practical line is now clear: ingestion is mostly legal, but recognizable verbatim output is where liability starts.
## Why the NYT vs. Brave Suit Matters for API Builders
The New York Times says Brave's crawler collected more than 30 billion web pages and the company sells that index through its search API to AI companies including Anthropic, Perplexity, Mistral, xAI, and Cohere. Many developers consume search APIs without knowing the provenance of the underlying data, which makes this case relevant beyond the named parties. The broader pattern on both sides of the Atlantic is the same: getting past a paywall turns reading news into taking it. If your app surfaces third-party search results, treating that content as safe requires knowing how it was obtained.
## The Rest of the Briefing: Power, Bounties, and Work Hours
OpenAI announced a Grid Protectors plan built on its Georgia contract, saying families and small businesses should not subsidize compute-campus electricity; Georgia Power estimates a typical home could save at least $15 per month from 2029. OpenAI is also offering up to $172,000 for the worst ChatGPT bugs, and new data suggests people who use AI at work are working longer days, not shorter ones. These stories are reminders that AI's operational costs and social effects are still being priced and negotiated.
For developers, the takeaway is to build with provenance: use licensed or clearly permitted sources, avoid paywall bypass, and filter near-verbatim generated text. The legal spotlight is moving from training data to the answers your application serves.
Build LLM apps with source provenance and no paywall-bypassing—today's rulings put output on the hook.