"Hiding in Plain Sight": When Language Becomes Hide-and-Seek Between Humans and AI
Imagine you and a good friend standing inside a transparent glass box, surrounded by a crowd of extremely intelligent yet extremely literal-minded "robot police."
Each robot holds a thick *Dictionary of Banned Words*. The moment you utter any word in that dictionary, sirens blare and you're thrown out of the box.
To keep talking, you and your friend have to invent a whole new communication system:
- You can't say "money," so you say "RMB" or "pocket change."
- You can't say "death," so you say "collected their final paycheck" or "became a star."
- You can't say "violence," so you say "physical persuasion."
- Modify too little, and AI catches you every time (evasion failure).
- Modify too much, and while AI can't catch you, your friends can't understand you either (communication failure).
As long as your friend understands you and the robots don't, you win. That is "algospeak."
In May 2026, a Stanford arXiv paper ("Algospeak, Hiding in the Open: The Trade-off Between Legible Meaning and Detection Avoidance") revealed a deep truth about language evolution: our language is turning into a game of hide-and-seek between humans and AI.
What Is Algospeak?
Author Jan Fillies argues that algospeak is a very special mode of communication. Unlike traditional "encryption," which aims to make messages unreadable to *everyone*, algospeak aims to "hide in the open" — it wants every human to understand instantly, while every AI moderator is left confused.
The Language "Tipping Point": The MUM Threshold
The paper's most striking finding is a tipping point called MUM (Majority Understandable Modulation).
Picture the dynamic process: 1. Light variant: You write "su1c1de" with substitutions. Humans spot it immediately — and so do AI systems via pattern matching. 2. Moderate variant: You switch to slang like "restart" or "remake." The latest AI moderators haven't learned the code yet, but human netizens get it instantly. 3. Heavy variant: You distort words beyond recognition, or use strings of seemingly meaningless memes as references.
A cruel trade-off exists here:
A Game With No End
Feynman once said: "What I find most interesting is that when you try to solve a problem, you often create new ones."
The world of algospeak is the same. It is a continuously spiraling evolutionary process: 1. Humans coin a new word. 2. AI learns it and adds it to the dictionary. 3. Humans discover the word no longer works, so they invent a "newer new word."
The paper notes that as large language models (LLMs) grow more capable, this evolution is accelerating exponentially. This is not just vocabulary drift — it is an offense-and-defense battle over "semantics" and "common sense." Humans are exploiting AI's lack of "subcultural context" and "real-time emotional connection" to build invisible defensive lines inside language.
Why It Matters
This is not just netizens fooling around on social media.
The paper reminds us that the very nature of language is changing. Language used to exist for "expression"; now, to some degree, it exists for "evasion."
When our habits of expression are shaped by algorithms, and when our dictionaries fill up with "unalive," "seggs," and assorted homophone puns, we are paying an invisible cost — the clarity of communication and the depth of thought are being eroded.
To summarize:
We whisper in broad daylight, building the digital world's underground railway out of codes only our own kind understands.
AI increasingly resembles that erudite but rigid old pedant: it guards all the rules but cannot grasp the winding curves of human emotion. Meanwhile we humans, through our endless "nonsense," preserve a sly vitality that algorithms cannot replicate.
The next time you see a baffling piece of internet slang, don't roll your eyes. It may not be posturing — it may be humans waging a tacit guerrilla war against AI to defend "the right to speak."
The most beautiful language is often that unspoken pact: "I get it, you get it, but it doesn't."