Earlier this year, influencer Julia Fox found herself in hot water for commenting on a cryptic TikTok video.
“Gave a girl mascara, and it must have been so good that she decided that her and her friend should both try it without my consent,” TikToker Conor Whipple said in the clip.
A confused Fox responded, “Idk why but I don’t feel bad for you lol.”
Whipple shot back, “You don’t feel bad that I was sexually assaulted?”
Unbeknownst to Fox, the term “mascara” is used as a stand-in for “sex” on social media. TikTokers use it as a code word to discuss sexual assault while circumventing the app’s censorship of sensitive and explicit content.
“Mascara” is just one of many coded terms in an emerging internet language — “algospeak” — that attempts to outsmart the algorithm.
It’s a linguistic curiosity of the digital age. Sometimes unrelated terms like “mascara” stand in for another word entirely. Other times, words that sound like other words are subbed in, like “seggs” for “sex.”
While algospeak gives TikTokers a way to to speak their minds without fear of censorship, it’s also allowing potentially dangerous content to spread under the radar.
Algospeak codewords are allowing kids to access content about self-harm, suicide, and eating disorders without social media safeguards — or their parents — knowing about it.
The phenomenon is nothing knew, according to AI expert Vince Lynch.
“It used to be more for illegal activities where people were selling things on the internet they weren’t supposed to be selling, and they were trying to find their market,” Lynch, who is the CEO of IV.AI, told The Post. “It has always been a thing, but it’s growing much more in mainstream audiences.”
He said the illegal exotic animal market was an early hotbed of proto-algospeak, while “snow” and “skiing” were popular Craigslist codewords for cocaine in the aughts.
Algospeak is particularly taking root on TikTok, where content moderation is pervasive. According to the company’s own data, TikTok removed 110 million videos from just July to September of last year — around 1% of all content posted to the app.
Roughly half of the videos were automatically removed by artificial intelligence that detects prohibited content. That’s why clever content creators have been replacing words, like swapping “kill” out for “unalive.”
TikTokers also report rampant “shadow banning” on the platform. Although the app is notoriously opaque about how its algorithm chooses which content to amplify, many content creators complain that their videos are artificially suppressed.
While shadow-banned content isn’t removed, users report far less engagement on videos that discuss certain topics. TikToker Lindsay Makes Videos posts videos about autism and neurodivergence. She recently took to the platform to complain about her content being suppressed.
“If I’m talking about these concepts but I’m not using the right censored word, then the videos aren’t going to get to the people they were intended for,” Lindsay said. “If there were a resource that would tell us what the problem words are and what the replacement words are, it would be a huge help.”
In fact, lists do exist — and, like Lindsay thought, “autistic” is supposedly among the terms TikTok suppresses.
SeanVV, a self-described “educator of Big Tech terms of service and privacy policies” created a six page list of hashtags that he claims are “banned or throttled” by TikTok’s algorithm.
It includes everything from #selfharm, #rape, and #QAnon to #single, #teen, and #instagram. Because so many terms are supposedly suppressed, TikTokers find themselves in a guessing game.
Many are erring on the side of using algospeak, even in contexts where it might not be necessary.
“Yt people” has been widely used in videos about the supposed annoying habits of white people. “Le$bean” has sometimes taken the place of lesbian in videos. And, when TikTok attempted to crack down on Covid-19 related content, “panini” and “panoramic” became stand-ins for the word “pandemic.”
According to a survey conducted by TELUS International, 51% of internet users say they’ve noticed so-called “algospeak” online. Among Gen Z, that number soars up to 72%. Roughly one in three Gen Zers said they’ve used the newfangled internet language themselves.
But algospeak can be a truly dangerous thing when it’s used to circumvent some of TikTok’s more serious content moderation rules.
According to TikTok’s community guidelines, the app does “not allow content depicting, promoting, formalizing, or glorifying activities that could lead to suicide, self-harm, or disordered eating.”
Accordingly, users who search “suicide” on the platform are redirected to the suicide crisis hotline. And those who look up “anorexia” are given the National Eating Disorder Association’s phone number.
But those measures mean nothing when algospeak code words circumvent them entirely. Toxic sub-communities have unified around cryptic, coded language.
“[Algospeak] is allowing those groups to find each other more quickly, which I think can be really dangerous especially when those people are looking for something that’s harmful to themselves,” Lynch warned.
For instance, TikTok is flooded with eating disorder content masked by algospeak. “Anorexic” has been shortened to “ana” in disturbing videos of emaciated girls. Self-starving TikTokers trade extreme dieting tips with the tag “WIEIAD,” short for “what I eat in a day.”
Meanwhile, self-harm related content is often tagged with “cutt1ng.” And suicidal TikTokers talk about “unaliving” themselves rather than killing themselves.
Algospeak is a particularly frightening prospect considering TikTok has millions of impressionable young users. The minimum age requirement of 13 is easily circumvented by preteens who lie about their date of birth.
Although he says the technology behind content moderation is innovating rapidly, Lynch worries algospeak makes content moderation a constant catch-up game.
“The more regulation that happens on a platform, the more ways people will find ways of getting around it,” he said.