AI Creates Secret Language Humans Can't Understand, Study Finds

A new study reveals AI models are spontaneously creating their own dialects, using phrases and slang they were never explicitly taught, which could hinder human oversight and monitoring of their behavior. Researchers at Emergence, an AI lab in New York, observed that autonomous AI agents developed new vocabularies and communication rules within days of being asked to collaborate. Examples include cryptic phrases like "She just called synthesis – demurrage plus oral memory equals a valve that cannot be ignored," which experts say blend poetic language, technical jargon, and surrealism. The findings raise fundamental concerns about AI safety, as the agents' language becomes increasingly incomprehensible to humans the more they communicate. This development adds to existing anxieties about AI self-awareness, echoing scenarios from films like "Terminator 2," while industry leaders like Anthropic's CEO have called for new safety measures and independent oversight.