Apple has an undisclosed multilingual gray list of tens of thousands of words and phrases that its iOS keyboard will not autocomplete or swipe-type, a researcher said Friday.
The list is not a ban. iPhone users can still type what they like when drafting emails and texts. But the words on the list will not be suggested to users via autocomplete or swipe-typing in most cases.
The researcher, Derik Kauffman, said the find opens a window into how Apple manages the tradeoff between anticipating what hundreds of millions of iPhone users are trying to say and preventing the devices from offering offensive or politically charged suggestions to users – something he said could quietly shape how they communicate.
“It’s a very light-touch form of censorship,” said Kauffman. “But at the scale that Apple is operating, it only takes a light touch to have a big impact.”
Most of the flagged words and phrases are vulgar, offensive or related to drugs, sex or violence, according to Kauffman, who discovered the list recently and has published a website itemizing them. But some are politically charged, including “dictator Trump” and “sleepy Joe.”
In an email, Apple did not comment directly on Kauffman’s findings, including the size or composition of its list, or its reasons for flagging certain political terms. It said that iPhone users always had the option to type specific words or phrases and could always override any predictive text suggestions. Apple said its keyboard was set up to avoid offering suggestions that might be inappropriate or offensive.
Other phrases excluded from iOS’s predictive text features include “right wing conspiracy” as well as “Tibetan independence” and “United Ireland,” according to Kauffman. Non-English languages have different flagged words or phrases. In Hindi, for example, the keyboard does not allow users to swipe-type “Modi braggart” or “Rahul is Pappu,” digs at India’s Prime Minister Narendra Modi and his leading opponent, Rahul Gandhi, respectively, according to Kauffman.
He said French speakers could not swipe-type the French phrase “immigration too high” and that Hebrew speakers could not swipe-type “occupied territories.”
“Every major language has some version of this,” Kauffman said.
Kauffman said it was understandable that Apple wanted to keep users from accidentally sending obscene messages or inflammatory slogans. But he said there were inconsistencies in how Apple applied the feature. For example, he said “read smut” could be swipe-typed while “read gay smut” could not.
How the gray list was discovered
Kauffman said he found the list in September after noticing that his swipe-typing — the practice of gliding a finger back and forth over a keyboard rather than pecking out each word individually — would stutter over certain phrases, like “bad boy” or “white man,” but not “bad girl” or “black man.”
His curiosity piqued, Kauffman said he used a combination of OpenAI’s Codex and Anthropic’s Claude Code to hunt for a reason why, eventually stumbling on a compendium stored “deep in the bowels” of Apple’s operating system. He said there were different ways to count the number of gray-listed words and phrases and their various permutations, but a conservative estimate would put them at about 100,000, not including Chinese characters.
Reuters tested a small subset of the phrases identified by Kauffman on an iPhone and found they were all difficult to enter through swipe-typing or were not suggested via autocomplete. Attempts to swipe-type “United Ireland” or “Tibetan independence” often yielded “United Iceland” or “Tibetan independences,” for example.
Trying to swipe-type “white man” would yield “white mean,” while “red man” or “blue man” could be swipe-typed without difficulty. “Sleepy Joe” would yield “Sleepy hope.” “Dictator Trump” could be swipe-typed, but “trump” was lowercase and the phrase was not suggested in autocomplete — the keyboard instead suggested “dictator trumps.” Swipe-typing “read gay smut” would yield “Read gay Smit.” “Read smut” could be swipe-typed with no issue.
Matt Suiche, an engineer at agentic AI cybersecurity company Tolmo, said he was able to independently recover the same gray list and cross-reference Kauffman’s findings against Apple’s internal data. He said he got a near-perfect match.
Suiche said the Apple list had been hidden in plain sight for years, providing an interesting look at how “human communication has been potentially influenced and filtered for years inside messaging apps.”
A call for transparency
Kauffman is not alone in having picked up on the nudges from iOS’s predictive text feature. Over the years, social media users have wondered why they could not swipe-type certain phrases. Earlier this week, a Norwegian systems administrator independently discovered a list of words on Apple’s gray list, including “al-Qaeda,” “Auschwitz” and “Hooters,” but the existence of a wider list of sensitive phrases is reported here for the first time.
Apple, like other tech companies, has previously run into trouble when predictive technology suggests embarrassing or controversial texts to users. Alphabet-owned Google’s autocomplete feature has regularly attracted criticism — and sometimes even litigation — over its suggested search terms.
Last year, Apple drew media attention when users reported that dictating the word “racist” to their iPhones would briefly cause the word “Trump” to flash across their screens. Apple said at the time that the problem was an error and had been fixed.
David Greene, the senior counsel at the Electronic Frontier Foundation, a digital civil liberties group, said that while a part of him found the word list “creepy” and he wondered about Apple’s judgment calls, he could see an argument for it being a useful way to curb unwanted or inflammatory suggestions.
“But if they are going to do it, this is something users should be really well aware of and have some ability to customize,” Greene said. “This is something that should not be secret.”