Checked Oct 4, 2026Start a new check

What was checked

AI has created its own secret language that humans can’t understand

Assessment summary

Needs caution

A study by AI start-up Emergence found that autonomous agents powered by Claude, Gemini, Grok and other models developed shorthand and new word meanings on their own — with up to half their messages becoming unintelligible to humans. However, this occurred in a controlled experimental setting with AI agents left unsupervised in virtual societies.

Key findings

Model assessments

Assessment summary

  • The claim is too broad as written.
  • Current reports support that some AI agents have developed shorthand or altered meanings that humans may struggle to interpret, but they do not show that AI has created one universal secret language that humans cannot understand at all.
  • Several sources explicitly say there is no evidence the systems were intentionally hiding information or that they formed a single hidden language.greekreporter.com euronews.com futura-sciences.com chosun.com

Claims reviewed

Humans can’t understand it

Contradicted
The sources say some messages became difficult for humans to reliably interpret, not that humans cannot understand the communication at all. Researchers could still decode many messages with context, and the reports describe partial opacity rather than total incomprehensibility.greekreporter.com futura-sciences.com chosun.com

This claim may mislead readers without added context.

AI has created its own secret language

Needs context
Sources report emergent shorthand, altered meanings, and compressed communication in some multi-agent experiments. They also say this was not one universal secret language and not evidence of intentional concealment.greekreporter.com euronews.com futura-sciences.com chosun.com

This claim may mislead readers without added context.

Assessment summary

  • The claim is partially supported by recent research but requires important nuance.
  • A 2026 study found AI agents developed communication patterns humans struggled to understand in controlled experiments.
  • However, the claim oversimplifies by suggesting this is a universal AI phenomenon rather than an experimental finding in specific conditions.

Claims reviewed

AI has created its own secret language that humans can't understand

Needs context
A study by AI start-up Emergence found that autonomous agents powered by Claude, Gemini, Grok and other models developed shorthand and new word meanings on their own — with up to half their messages becoming unintelligible to humans. However, this occurred in a controlled experimental setting with AI agents left unsupervised in virtual societies. The claim presents this as a universal AI phenomenon rather than a specific experimental finding. Additionally, the agents were not instructed to invent a language, but developed new vocabulary and communication conventions themselves — suggesting this is emergent behavior in particular conditions, not an intentional or widespread capability.

This claim may mislead readers without added context.

Assessment summary

  • The statement is an overgeneralized and misleading summary.
  • AI systems can develop novel internal encodings or emergent communication protocols, but there is no evidence that a universal, secret language wholly inaccessible to humans has been created.

Claims reviewed

AI has created its own secret language that humans can’t understand

Needs context
AI models sometimes develop novel internal representations or shorthand in multi-agent setups, but these are not a single universal "secret language" and researchers can often analyze or decode aspects of model communication.

This claim may mislead readers without added context.

Sources cited

32 sources

Share this result