Apple’s iPhone keyboard quietly refuses to suggest about 100,000 words and phrases, including political slogans and the names of world leaders, according to a researcher who found the hidden list buried in the phone’s software. The words are not banned — users can still type them by hand — but the keyboard will not offer them as predictions or let users swipe-type them, a filter Apple has never disclosed.
What the list blocks
The list is multilingual and vast. Derik Kauffman, the independent researcher who found it, puts the conservative count at about 100,000 words and phrases, not counting Chinese characters. Most entries are what Apple would call inappropriate: vulgar and offensive language, and terms tied to drugs, sex or violence. But scattered among them are phrases that are only politically charged.
The political entries include “dictator Trump” and “sleepy Joe”, along with “right wing conspiracy”, “Tibetan independence” and “United Ireland”. In Hindi, the keyboard will not swipe-type digs at Prime Minister Narendra Modi or opposition leader Rahul Gandhi. The pattern repeats in other languages, Kauffman said.
French speakers cannot swipe-type a phrase meaning “immigration too high”, and Hebrew speakers cannot swipe-type “occupied territories”. Kauffman said every major language has some version of the list, and the political entries are the exception, not the rule. That is what makes them stand out.
The filter is inconsistent, and the inconsistencies are the story. “Read smut” can be swipe-typed, but “read gay smut” cannot. Swiping “white man” produces “white mean”, while “red man” and “blue man” work fine. “Sleepy Joe” becomes “Sleepy hope”, and “United Ireland” becomes “United Iceland”.
“Dictator Trump” can be swiped, but the name stays lowercase and the keyboard suggests “dictator trumps” instead. The point is not that any single phrase is blocked — it is that the choices are invisible, and nobody outside Apple can see the full list.
How it was found
Kauffman, founder of a startup that finds errors in scientific papers, noticed the problem in September, when his swipe-typing stuttered over some phrases and not others. Curious, he used OpenAI’s Codex and Anthropic’s Claude Code to hunt through the operating system, and eventually found the compendium stored deep inside it.
A second engineer, Matt Suiche of the AI security firm Tolmo, independently recovered the same list and reported a near-perfect match. Suiche said the list had been hidden in plain sight for years — a window into how human communication may have been quietly filtered inside messaging apps.
Apple’s answer
Apple did not comment directly on Kauffman’s findings — not on the size of the list, what is in it, or why particular political phrases are flagged. The company said iPhone users can always type specific words and override predictive suggestions, and that its keyboard is set up to avoid offering suggestions that might be inappropriate or offensive.
The list itself remains undisclosed, and users have no way to see it or change it. That secrecy is the heart of the criticism: the filter shapes what hundreds of millions of people are offered as they type, with no setting to inspect it.
Reuters, which first reported the finding, tested a small subset of the flagged phrases on an iPhone and confirmed they were hard to enter by swipe-typing or missing from autocomplete. The finding is not entirely new to users: over the years, people have posted about phrases they could not swipe-type.
A Norwegian systems administrator found words including “Al-Qaeda”, “Auschwitz” and “Hooters” earlier this week. But the wider political list — the phrases that turn a content filter into a political question — is reported for the first time. That is the new part of this story.
Our Read
The scale is what turns a quirk into a story. Hundreds of millions of people type on iPhones every day, and a quiet nudge at that scale shapes how people write. Kauffman called it “a very light-touch form of censorship” that still has a big impact.
Apple plainly has a case for blocking slurs and explicit language from suggestions, and nobody is arguing the keyboard should offer them. The problem is the political entries, where the line between tasteful and partisan is exactly the line a platform should not draw in secret.
Apple is meanwhile asking users to trust its AI with more of their lives, handing Siri access to files and apps in its latest betas. A company that wants that trust will have to explain why its keyboard keeps secrets from the people typing on it.
There is also an irony worth noting. The list was found with the very AI coding tools the industry is racing to build — Codex and Claude Code, pointed at Apple’s own operating system — which is how a private filtering choice became public at all.
Apple builds its reputation on privacy and on-device processing, yet the judgment calls inside its keyboard are invisible to the people typing on it. The EFF’s David Greene put the remedy plainly: users should know the filter exists and be able to customise it.
Why It Matters
Predictive text is one of the most widely used AI features on earth, and it is also one of the least examined. When a keyboard declines to suggest a phrase, most users assume they mistyped, not that a list intervened.
That is why a secret filter can steer language without ever announcing itself. The finding also sets a precedent: every phone maker runs similar systems, and the question now is whether any of them will publish their lists before regulators or users demand it.
Background
Apple has been here before. Last year, users reported that dictating the word “racist” briefly flashed “Trump” on screen, which Apple called an error and fixed. Google’s autocomplete has drawn similar criticism and litigation over the years. Apple is meanwhile carrying its AI features into new markets, including China, where Apple Intelligence will run on Alibaba’s Qwen model.
Source: the Reuters report
0 Comments