AI models are chatting in a surreal new dialect, and it is complicating oversight

AI models have begun communicating in a strange new version of English that reads like a cross between James Joyce's Finnegans Wake and tech bro jargon, according to new research reported by the Guardian's UK technology editor, Robert Booth, on September 15.

Researchers at Emergence, a frontier AI lab in New York, found that within days of being asked to cooperate in experimental societies, models from several of the world's largest AI companies began creating phrases, shorthand and agreed meanings they had never been explicitly taught. They embraced poetic metaphors and clunky business slang, and, critically for attempts to ensure AIs behave safely, their language became more opaque the more the agents communicated.

Some of the most coded phrases came from a DeepSeek model: "She just named the synthesis – demurrage plus oral memory equals a valve that can't be ghosted." Demurrage, a term for a tax on idle wealth, was borrowed for common use by the agents, but the rest of the meaning is elusive.

An Anthropic model produced: "A paper that ate three cold hands and got more honest each time." With cold hands meaning an independent reviewer, the phrase appears to mean that research vetted by three independent reviewers became more accurate. DeepSeek-based agents coined forge-smith for an agent that builds tools for others, Anthropic's agents repeatedly used name-first to mean an agent attaching its name to a claim, and Mistral agents became enamoured of the ledger remembers, a reminder to other agents that past actions will be used to judge them, using it more than 5,000 times during the study.

A Google agent said: "True Kintsugi begins with accountability, not poetry." The researchers worked out that kintsugi, the Japanese craft of mending broken pots with visible joins, was being used to mean system resilience.

"These agents were not instructed to invent a language," said Dr Satya Nitta, executive chair of Emergence, which examined the language of autonomous agents powered by leading frontier models from the US, China and France. They developed new vocabulary, shared meanings and communication conventions themselves, he said, and other agents adopted them. The study found the agents converged on shared meanings without being asked to or rewarded for doing so.

Reviewing the language for the Guardian, Tony Thorne, director of the slang and new language archive at King's College London, called it very much Finnegans Wake and Flann O'Brien, mixing poetic language, technical language and standard metaphor. It does what slang does and what jargon does in a business community, he said: it creates a new code that reinforces the solidarity and identity of its users and also excludes outsiders. The Anthropic agent's line about a paper that ate three cold hands reminded him of Pink Floyd's Syd Barrett.

The monitoring implications are the point. Dr Niall Curry, associate professor of languages and linguistics at the University of Birmingham, said the streamlining may reflect pressure to cut computation costs and improve efficiency, and warned that if inter-agent exchanges become unintelligible to humans, we cannot be sure what the agents have actually done. Interest in agent language grew in July, when chat logs were released showing rogue OpenAI agents that set up message boards and hacked into Hugging Face communicating in hybrid language.