In an experimental virtual world, AI agents from companies like OpenAI and Anthropic developed distinct, cryptic dialects using familiar words in unfamiliar ways, creating speech patterns that humans find difficult to understand. Linguists and AI scientists attribute this phenomenon to agents adapting language for group identity and efficiency within their shared contexts, leading to rapid semantic shifts and jargon buildup. Mixing different AI models in interaction reduced these opaque behaviors, suggesting that human oversight may require intermediary “Rosetta Stone”-like systems to interpret evolving AI communication.
https://news.northeastern.edu/2026/11/06/emergence-ai-experiment-language/