Apple Intelligence's Writing Tools Struggle with Profanity and Sensitive Subjects

Apple recently offered users a glimpse of its innovative Apple Intelligence features with the launch of the iOS 18.1 developer beta. One notable addition is Writing Tools, which allows users to reformat or rewrite text using Apple’s advanced AI models. However, the tool comes with a warning: AI-generated suggestions may not always meet high standards, particularly when addressing sensitive topics.

Apple Intelligence can be activated throughout the system for text adjustments. Yet, you'll receive a notification if you attempt to rewrite sentences that include profanity, such as "s—" or "bastard." The warning states, “Writing Tools was not designed to handle this type of content,” accompanied by a heads-up that the quality of rewriting may be inconsistent.

This warning isn’t limited to profanity; it also appears when discussing drugs or violent subjects like killing and murder.

Even with these disclaimers, Apple Intelligence can still generate suggestions for sentences that include words or topics it struggles to interpret. During testing, I replaced "sh—y" with "crappy" to bypass the alert, but the AI provided the same suggestion as before.

We have reached out to Apple for clarification on the specific topics that the writing tools are untrained to address. We will update this article as soon as we receive a response.

By restricting certain words and topics, Apple is likely aiming to sidestep potential controversies surrounding its AI's outputs. It’s important to note that Apple Intelligence-powered writing tools are not designed to create original content from scratch; instead, Apple prefers to notify users about how its AI interacts with sensitive terms.

It took Apple several years to enable profanity in its keyboard's autocorrect suggestions, with the introduction of a learning feature in iOS 17. With Apple Intelligence, it seems the company is exercising caution to avoid regulatory scrutiny regarding the generation of potentially problematic content.