Users have flooded social media with complaints that iPhone autocorrect has suddenly become erratic, changing words like 'come' to 'coke' and 'winter' to 'w Inter'. A video showing the keyboard altering 'thumb' to 'thjmb' has amassed over 9 million views since iOS 26 launched in September.
Experts say the root cause may be Apple's shift to an AI-powered transformer language model, a more advanced system than the 'n-gram' technology used for decades. Statistician Jan Pedersen, who worked on early autocorrect at Microsoft, noted that these models are 'far harder to understand when things go wrong'.
Kenneth Church, a computational linguist who pioneered autocorrect in the 1990s, added: 'What Apple does is always a deep, dark secret.' Apple declined to elaborate on the new model but confirmed the video issue was unrelated to autocorrect.
Autocorrect evolved from simple spellchecking in the 1970s, which just compared words against a dictionary. Modern autocorrect must statistically determine context, such as whether 'graff' meant 'giraffe' or 'graph'. The new transformer model, similar to those powering ChatGPT, learns patterns on-device but remains opaque to users.
