Over 100,000 hidden autocomplete blocks uncovered inside iOS keyboard

A researcher has surfaced a hidden blocklist of more than 100,000 words and phrases that Apple's iOS keyboard will not offer as autocomplete suggestions, publishing the full inventory online and raising questions about how quietly the company filters what users type.

According to The Verge, which cites original reporting by Reuters, researcher Derik Kauffman extracted the list from iOS and found that it covers phrases in all of the world's major languages.

How the filtering works

The blocks do not prevent anyone from typing a term outright. As The Verge explains, a blocked phrase can still be entered character by character — it simply never appears as a predictive suggestion, and the keyboard may rewrite it when a user types by swiping. In one example Kauffman documented, swiping out "left wing conspiracy" produced "left wrong conspiracy."

The bulk of the list is unremarkable from a moderation standpoint: slurs, vulgarities and terms related to sex and drugs. But the collection also contains politically charged entries such as "illegal immigrant," "dictator Trump," "sleepy Joe" and "covid was a hoax," alongside comparatively harmless strings like "Santa isn't real" and "iMessage isn't working."

Kauffman also noticed inconsistencies. Some phrases are blocked while their mirror images are not — "bad girl" is suppressed, for instance, while "bad boy" is left alone. He has published the complete list, including separate breakdowns for individual languages.

How the list was found

Kauffman told Reuters he used OpenAI's Codex and Anthropic's Claude Code to help locate the data, which sat buried deep inside iOS. On his website, he describes the phrases being stored as a hash table — a structure that obscures the raw text — which he converted back into readable words. A second researcher told Reuters they independently pulled the same list from the system, which supports the finding's accuracy.

Apple's position

Asked about the list, Apple told Reuters that its keyboard is built to avoid surfacing potentially inappropriate or offensive suggestions, while emphasizing that iPhone owners can still type whatever they want or dismiss the predictions they are given. The company did not immediately respond to The Verge's request for further comment.

Why it matters

Autocomplete is not a neutral convenience; it shapes what billions of keystrokes look like, and until now the list governing iOS suggestions was invisible — stored in hashed form, absent from any policy page, and never disclosed by Apple. That matters for two reasons.

First, the list reaches well beyond obvious profanity into political territory, and it does so asymmetrically. Decisions about which political phrases deserve suppression — and which counterparts do not — were made privately, with no published criteria and no avenue for correction.

Second, the swipe-typing behavior is more consequential than withholding a suggestion. Silently changing what a user intended to write crosses from filtering a recommendation into altering input, and most people would have no way to notice it happening or understand why.

The episode also signals a shift in the balance of power around proprietary software. Systems that were once effectively unauditable can now be reverse-engineered with AI coding assistants doing part of the digging. Companies that embed hidden blocklists in mass-market products should expect those lists to come to light eventually — along with scrutiny of the gap between "we avoid offensive suggestions" and "we quietly edit your swipes."