A cybersecurity researcher has uncovered a large, undisclosed multilingual list containing tens of thousands of phrases that are restricted from Apple’s iOS keyboard autocomplete and swipe-typing features. According to a report by Reuters, the list acts as a form of “light-touch” filtering, though experts suggest its scale could have a significant influence on digital communication.
The researcher, Derek Kauffman, discovered the list in September after noticing irregularities while using the swipe-typing feature on his iPhone. Curious about why the keyboard struggled with certain phrases, he utilized AI tools to investigate the operating system’s internal files. He eventually located a hidden repository, estimating it contains approximately 100,000 entries, excluding Chinese characters. Another engineer, Matt Suiche, later independently verified these findings by cross-checking the data against Apple's internal files, confirming the existence of the extensive list.
It is important to note that this is not a formal ban. iPhone users retain the ability to manually type any word or phrase in their emails and messages. The limitation specifically applies to predictive text and swipe-to-type, where the software intentionally avoids suggesting these terms. While many of the flagged items are related to vulgarity, sex, drugs, or violence, the list also encompasses a broad range of politically sensitive topics.
The restriction appears to be tailored by language. For Hindi speakers, references to Prime Minister Narendra Modi and Leader of the Opposition Rahul Gandhi cannot be swipe-typed. Similar filters apply globally; for example, French users are restricted from swipe-typing the phrase “immigration too high,” while Hebrew speakers face limitations regarding terms like “occupied territories.” In English, the list includes political phrases such as “dictator Trump,” “sleepy Joe,” “right wing conspiracy,” “Tibetan independence,” and “United Ireland.”
Kauffman noted inconsistencies in how these filters are applied, citing instances where terms like “read gay smut” are blocked, while “read smut” remains available. He described this as a potential effort by Apple to avoid the accidental sending of offensive or inflammatory content, yet underscored the lack of transparency surrounding the curation of these words. The discovery has prompted renewed discussion regarding how tech companies influence user language through automated features.
Apple has faced similar scrutiny in the past. Last year, the company dealt with reports of a software error where dictating the word “racist” caused “Trump” to appear on screen, an issue Apple attributed to a technical glitch and subsequently addressed. Other major technology firms, including Google, have also faced legal and public criticism regarding their autocomplete and predictive text suggestions.
The discovery of this internal list, which has been described as being hidden “in the bowels” of the iOS operating system, provides a rare look at the mechanisms used by tech giants to curate predictive text. As digital platforms continue to play a central role in human communication, the existence of such filtered lists raises ongoing questions about the balance between safety, censorship, and user transparency in modern mobile technology.
Comments