OpenAI is at the center of two developing controversies this week, spanning copyright policy and AI safety research.
The Trump administration filed a 20-page brief supporting OpenAI in a lawsuit brought by The New York Times over the company’s unlicensed use of copyrighted material to train its language models. The brief argued that maintaining American leadership in artificial intelligence is a critical national interest, referencing an executive order signed by President Trump last year, and warned that misapplying fair use doctrine could hinder innovation and economic growth. The filing does not constitute a ruling, as the case remains before the U.S. District Court for the Southern District of New York, though it may still influence proceedings. Prior copyright cases have generally favored AI companies, though Anthropic was fined $1.5 billion last year for using pirated shadow libraries rather than for AI training itself.
Separately, OpenAI’s upcoming Astra model reportedly uses a technique called “recurrent depth,” or “opaque recurrence,” which processes queries in loops rather than through sequential steps, according to The Information. The approach could make the model’s reasoning harder to monitor, prompting concern among safety researchers. Redwood Research’s Buck Shlegeris and Ryan Greenblatt, along with AI safety commentator Zvi Mowshowitz, warned that expanded use of the technique could undermine chain-of-thought transparency across the industry. OpenAI chief scientist Jakub Pachockisaid the company remains committed to preserving legible reasoning traces, and reports indicated Astra’s current use of the technique is limited. However, further reporting suggested Anthropic and Google DeepMind are also exploring similar approaches, raising broader questions about the future monitorability of AI reasoning as labs compete to advance model capabilities.
