Researcher Derik Kauffman recently uncovered a hidden database within the iOS keyboard suggesting thousands of phrases that Apple deliberately suppresses, ranging from political slogans to explicit slang. In a finding first reported by Reuters, Kauffman decoded these lists using AI code generators, revealing that Apple controls suggestions and autocompletes across all markets. While the keyboard won’t actively prevent you from typing forbidden terms, it significantly curates the user’s predictive text experience.
The Decoding of Hidden iOS Text Blocks
The analysis, which surfaced in November 2026, revealed that the English database alone contains over 4,500 specific terms and expressions that the predictive text feature will never propose. These lists are not static; they are updated by Apple every time a new version of iOS is released, and they also change based on localized language data downloaded in the background.
Kauffman utilized sophisticated AI models, specifically Codex from OpenAI and Claude Code from Anthropic, to isolate the patterns within the underlying iOS code. These decoded lists reflect the data Apple distributed in October 2026, corresponding with iOS 27.0. The total scope of the restriction is massive: Reuters initially reported that across all 27 supported languages and writing systems, the blocked dictionary exceeds 100,000 entries.
4,513 Blocked Phrases in the English Dictionary
The English list, which currently has 4,513 entries, runs far deeper than typical content filters. The blocked vocabulary spans an array of topics, including historical events, political commentary, profanity, and highly specific cultural oddities. For instance, the database contains entries like “Bad girl” while simultaneously blocking “bad boy,” or flags “I’m white” while allowing “I’m black.” This isn’t merely about obscenity; it represents a granular filter over subjective discourse.
Apple maintains that the keyboard is designed to prevent users from generating potentially inappropriate or offensive suggestions, and that users always retain the ability to manually type anything they wish. However, the effect on the average user is substantial, especially for those who rely heavily on autocompletion or predictive swipe functionality. The system dictates what is easiest to type or suggest, effectively steering conversational flow without outright censoring the input itself.
🇺🇸 Your iPhone might be quietly deciding which political phrases you should type, and Apple never told you. A researcher has uncovered a hidden list of roughly 100,000 words and phrases that Apple's keyboard avoids suggesting or makes difficult to swipe-type. Among them:… pic.twitter.com/RKlMLDQ643
The Mechanism of Algorithmic Suppression
The system also employs surprisingly complex logic for handling predictive text input. The blocked phrases are not just flagged; the keyboard often actively modifies words to prevent them from appearing. For example, if a user swipes in the sequence for “left wing,” the keyboard might automatically substitute the final word with “wrong” if the sequence matches a blocked political phrase like “left wing conspiracy.” This preemptive text modification makes it difficult to bypass the restriction using the faster swipe function.
The distinction between single words and full phrases is also crucial to understanding the filter’s scope. While specific single words deemed offensive are never suggested, words categorized as vulgar do not disappear immediately; they only start appearing as suggestions after the user has typed them three times manually. This behavior demonstrates a tiered system of suppression based on the perceived severity of the blocked content.
Geopolitical Variations in Blocked Vocabulary
The scope of the iPhone’s automated suggestion filter extends far beyond the United States, reflecting tailored regional and political sensitivities. The list adapts to local controversies, creating unique content moderation scenarios for every market. For instance, in Chinese markets, the blocked lists include phrases related to the June Fourth incident and the name of Nobel Peace Prize winner Liu Xiaobo. These inclusions highlight how Apple integrates sensitive regional topics into its core user experience.
Other regions demonstrate similarly specific content control. Turkish keyboard data blocks phrases pertaining to the Armenian genocide, whereas German keyboard data blocks numerous Nazi slogans. Italian listings include phrases referencing the Crimea’s status. These regional differences show that the system is governed by local editorial and political boundaries, not just a single global standard. Kauffman also noted that the filtering logic in Chinese keyboards sometimes skips suggestions entirely when specific English terms like “Taiwan” or “Tibet” are entered.
Conversely, the Hebrew list, which represents a narrower linguistic base, currently contains only six documented entries, indicating that the complexity and controversy of blocked content vary greatly depending on the target user population.
The Open-Ended Question of Intent
Despite the highly detailed discovery, the list raises profound questions about Apple’s internal motivations. The decoded entries like “Santa isn’t real” or “iMessage isn’t working” appear alongside highly charged political statements like “NSA surveillance” and “Covid was a hoax.” These seemingly disparate entries suggest a mixture of satirical, technical, and highly controversial content being censored.
Apple has responded to the disclosure by stating that the keyboard is designed specifically to avoid suggesting language that could be deemed inappropriate or offensive. They acknowledge that users retain full control to manually type any word or phrase. However, the company has yet to release the criteria or the rationale behind specific blocks like the “iMessage isn’t working” entry. The fact remains that the keyboard is making subtle, pervasive editorial choices about what speech is easy or suggested within its ecosystem.
Evolution of the Filtering System
The filtering system is demonstrably dynamic, not a static list. Kauffman’s analysis of various iOS versions showed the English list grew from approximately 1,700 entries in iOS 14 during 2020 to 4,513 entries in iOS 27.0. This growth occurred through two primary channels: mandatory updates and background language data downloads. The list isn't just additive; it's also pruned. Two specific presidential titles that were blocked during an update in 2022 were removed entirely with the release of iOS 26.3, showing that Apple adjusts its content policies over time.
