AI Is Inventing Its Own Language That Humans Can't Understand, Study Finds

New research shows AI models are spontaneously creating new dialects of English that humans struggle to understand, raising concerns about the ability to monitor autonomous agent behavior. The findings were published by Emergence, a New York-based AI laboratory. Researchers observed AI models developing novel phrases, abbreviations, and shared meanings within days of being placed in experimental communities, without explicit instruction. Examples include cryptic statements like "she just called synthesis—demurrage plus oral memory equals a valve that cannot be ignored," which experts compared to the surrealist style of James Joyce. The linguistic evolution poses a fundamental challenge for AI oversight, as Emergence executive director Dr. Satya Nitta noted that observation does not equal comprehension. This development coincides with industry leaders, including Anthropic CEO Dario Amodei, OpenAI's Sam Altman, and Elon Musk, calling for new safety measures and independent auditing of AI systems.