Hardware Ready Ready to Bench?
iOS autocomplete. A researcher discovered an undisclosed list of approximately 100,000 words and phrases in various languages that are excluded from iOS autocomplete and swipe typing functions. The block does not prevent users from manually typing the terms, but it includes expressions considered offensive or politically sensitive.
According to Reuters, the iOS keyboard uses a hidden multilingual list that excludes certain words and phrases from automatic suggestions and swipe typing. Researcher Derik Kauffman identified this list and published the complete inventory on swipeblocked.com, categorized by language.
iOS autocomplete: why it matters
This is not an absolute ban: users can still type any term manually or ignore predictive suggestions. However, expressions in the list do not appear among the proposals and are automatically modified by the keyboard during swipe typing.
Kauffman estimates that the list comprises approximately 100,000 entries, excluding Chinese characters, noting that the count may vary depending on the permutations considered. Most of the terms are vulgar, offensive, or related to sex, drugs, and violence.
The Italian list currently includes over 306 terms, including expressions such as “Saudi Arabia Calabria,” “Crimea Russian state,” and “leftovers from prison.” In the United States, politically charged phrases appear, such as “illegal immigrant,” “dictator Trump,” “sleepy Joe,” “covid was a hoax,” “left wing conspiracy,” and “right wing conspiracy,” along with “Tibetan independence” and “United Ireland.” Among the less controversial cases are “Santa isn’t real” and “iMessage isn’t working.”
The list varies by language. In Hindi, according to Kauffman, it is not possible to swipe-type derogatory expressions directed at Indian Prime Minister Narendra Modi and his main opponent Rahul Gandhi. In French, the phrase “immigration too high” is blocked, while in Hebrew, “occupied territories” is restricted. The researcher stated that every major language has a similar version of this system.
Kauffman noticed the issue in September when swiping would “stumble” on phrases like “bad boy” or “white man,” but not on “bad girl” or “black man.” He used OpenAI’s Codex and Anthropic’s Claude Code to locate the list within the operating system, which was stored as a hash table and then converted back into text.
Reuters verified a subset of the reported phrases on an iPhone, confirming that they were difficult to enter via swipe or absent from suggestions. A second researcher, Matt Suiche, an engineer at cybersecurity firm Tolmo, independently recovered the same list and compared it with Apple’s internal data, obtaining a near-perfect match.
According to Suiche, the list has remained “in plain sight” for years and demonstrates how human communication has potentially been influenced and filtered within messaging apps. Kauffman believes this discovery highlights the balance between predicting what users want to write and the need to avoid offensive or politically charged suggestions, with a possible impact on how people communicate.
What changes and what are the effects
Kauffman described this system as a “very light form of censorship,” emphasizing that given Apple’s scale, even a small intervention can have a significant impact. While understanding the intent to avoid accidentally sending obscene messages or inflammatory slogans, he contests its consistent application.
Apple did not directly comment on the findings, nor on the size or composition of the list, nor on why certain political terms were included. Via email, it reminded users that they can always type desired words and ignore predictive suggestions, emphasizing that the keyboard is designed to avoid proposing potentially inappropriate or offensive content.
David Greene, senior counsel at the Electronic Frontier Foundation, described the list as partly “unsettling” and questioned Apple’s judgments, while acknowledging it could be a useful tool for limiting unwanted or provocative suggestions. He added that users should be aware of this feature and have the ability to customize it.
Similar predictive technologies have already caused problems for other companies, such as Google, which has been criticized and sued for suggested search terms. Last year, Apple was in the spotlight when some users reported that dictating the word “racist” briefly displayed “Trump” on screen: the company explained it was an error, later corrected.
Source and further reading on iOS autocomplete: original article.
* Content created with the assistance of artificial intelligence systems.