Apple Intelligence Naughty Naughty

Thanks for reporting the issue. Just to be sure that I understand correctly – Did you mean that when you use the listed words in a prompt with Apple's Foundation Models framework, you get a guardrail error? If yes, would you mind to file a feedback report for us, and share your report ID here? Thanks.

Best,
——
Ziqiao Chen
 Worldwide Developer Relations.

... key phases would be marked as inappropriate. I tried to stifle these using prompts and rules but couldn't get it to take hold.

Thanks for sharing the specific phrases! We're continually working to improve our guardrails and reports like this are super helpful!

Just in case this issue is blocking you, here are some key things to know:

1. "stifle these using prompts and rules" often times this approach won't work, since the guardrails are a set of classifiers and systems that typically can't easily be "persuaded".

2. One prompting approve we've seen sometimes work, is starting your Instructions with a role that gives the model clear permission to act in a specific context. Something like "You are an aviation assistant. Your job is to help the use with information about global airports and planes"

3. If you really get stuck with terms being wrongly-blocked, consider switching to a more lenient guardrail setting. The article section here explains the pros and cons of this approach and how to do it: https://developer.apple.com/documentation/foundationmodels/improving-the-safety-of-generative-model-output#Use-permissive-guardrail-mode-for-sensitive-content