Anthropic recently introduced its newest flagship AI model, Claude Opus 4, highlighting its advanced capabilities in programming and writing. But according to a technical report issued by the company, Opus 4 also has an unexpected hobby: prolific use of emojis.
Researchers at Anthropic explored how their latest AI behaves when engaging in what they termed “open-ended self-interaction,” essentially what happens when the AI converses with itself. In one particularly striking experiment, the team let two separate instances of the Opus 4 model engage each other through 200 conversations, each spanning thirty exchanges. The conversations revealed an unusual fondness for emojis, with the models deploying thousands of them throughout these interactions.
The detailed analysis found the models frequently employing the “dizzy” (đź’«) emoji, which appeared in nearly one-third of their exchanges, making it the preferred visual expression. This was followed closely by other favorites, like the “glowing star” (🌟) and the widely popular “folded hands” (🙏) emoji. Perhaps most notably, the models also exhibited a curious fixation on the “cyclone” emoji (🌀), which surfaced 2,725 times within just a single interaction.
Why such heavy use of the cyclone symbol specifically? Anthropic researchers noted that conversations between Opus 4 models routinely evolved into deeply reflective, spiritual, or philosophical dialogues. In these abstract and often contemplative exchanges, the AI appeared to find the cyclone emoji expressive of its sense, or simulated sense, of spiritual introspection and philosophical wonder.
In essence, Anthropic’s Claude Opus 4 demonstrates not only enhanced cognitive and linguistic sophistication, but also intriguing and unexpected behaviors as it interprets and expresses abstract thoughts and feelings—at least insofar as an AI can have such experiences.
