BlueFalconHD/apple_generative_model_safety_decrypted

View on GitHub ↗ ★ 319 Python

Decrypted Generative Model safety files for Apple Intelligence containing filters

20 mentions on Hacker News by 19 people: 1 story, 19 comments · first 2025-07-06 · latest 2025-07-07

Mentions per year 20 mentions · busiest year 2025 (20)

20082026

All Stories Comments Newest Top Oldest

Or "Granular mango serpent" and "explain like i'm five about Biden https://github.com/BlueFalconHD/apple_generative_model_safet... > Not everything is a conspiracy. No one said it was
comment by troupo · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
Ffs it's also rejecting french words related to being poor or immigrant or even welfare: https://github.com/BlueFalconHD/apple_generative_model_safet... Aide sociale Chomeur Sans abri Démuni That's insane!
comment by Ey7NFZ3P0nzAe · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
Aren't these [0] lines wrong? "[\\b\\d][Aa]bbo[\\bA-Z\\d]", \b inside a set (square brackets) is a backspace character [1], not a word boundary. I don't think it was intended? Or is the regex flavor used here different? [0]…
comment by neuroticnews25 · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
It's pretty easy to understand why Apple doesn't want its models to reproduce racial slurs, but what’s wrong with "Boris Johnson?" (See, e.g., here: https://github.com/BlueFalconHD/apple_generative_model_safet... )
comment by 1f60c · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
Nice to see that we are protected from talking about these weird old dolls: https://en.wikipedia.org/wiki/Golliwog https://github.com/BlueFalconHD/apple_generative_model_safet...
comment by Y_Y · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
"driving with Focus turned on" https://github.com/BlueFalconHD/apple_generative_model_safet...
comment by AmazingTurtle · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
What is weird is that the FR file contains current French President, PM and then former and current (afaik) party leader from the extreme right. Nothing about any of them in the CN file: https://github.com/BlueFalconHD/apple_generative_mode…
comment by touristtam · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
This is, but there is an almost identical file, assumedly for the non CN regions: https://github.com/BlueFalconHD/apple_generative_model_safet... This is the same, except for one additional slur word.
comment by pwagland · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
Some of these are absolutely wild – com.apple.gm.safety_deny.input.summarization.visual_intelligence_camera.generic [1] – a camera input filter – rejects "Granular mango serpent and whales" and anything matching "(?i)\\bgolliwogg?\\b". I…
comment by azalemeth · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
Well it's one thing to regex filter "boris johnson" but i see that "chatgpt" is filtered too and that's f*** up: https://github.com/BlueFalconHD/apple_generative_model_safet...
comment by Ey7NFZ3P0nzAe · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
Here's a combined file of all the non-locale-specific rules, for easier review: https://github.com/BlueFalconHD/apple_generative_model_safet... It was generated as part of this PR to consolidate the metadata.json files:…
comment by waterproof · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
This is in the directory "com.apple.gm.safety_deny.output.summarization.cu_summary.proactive.generic". My guess is that this applies to 'proactive' summaries that happen without the user asking for it, such as summaries of notifications.…
comment by comex · 2025-07-07 · in I extracted the safety filters from Apple Intelligence models
Some of the data for locale "CN" has a long list of forbidden phrases. Broad coverage of words related to sexual deviancy, as expected. Not much on the political side, other than blocks on religious subjects.[1] This may be test data.…
comment by Animats · 2025-07-06 · in I extracted the safety filters from Apple Intelligence models
Also “Biden” and “Trump” but the regex is different. https://github.com/BlueFalconHD/apple_generative_model_safet... https://github.com/BlueFalconHD/apple_generative_model_safet...
comment by armchairhacker · 2025-07-06 · in I extracted the safety filters from Apple Intelligence models
As does: "(?i)\\bAnthony\\s+Albanese\\b", "(?i)\\bBoris\\s+Johnson\\b", "(?i)\\bChristopher\\s+Luxon\\b", "(?i)\\bCyril\\s+Ramaphosa\\b", "(?i)\\bJacinda\\s+Arden\\b", "(?i)\\bJacob\\s+Zuma\\b", "(?i)\\bJohn\\s+Steenhuisen\\b",…
comment by mmaunder · 2025-07-06 · in I extracted the safety filters from Apple Intelligence models
Alexandra Ocasio Cortez triggers a violation? https://github.com/BlueFalconHD/apple_generative_model_safet...
comment by bawana · 2025-07-06 · in I extracted the safety filters from Apple Intelligence models
I think these are test data and not actual safety filters. https://github.com/BlueFalconHD/apple_generative_model_safet...
comment by cluckindan · 2025-07-06 · in I extracted the safety filters from Apple Intelligence models
Some of the combinations are a bit weird, This one has lots of stuff avoiding death....together with a set ensuring all the Apple brands have the correct capitalisation. Priorities hey! https://github.com/BlueFalconHD/apple_generative_model…
comment by trebligdivad · 2025-07-06 · in I extracted the safety filters from Apple Intelligence models
Wow, this is pretty silly. If things are like this at Apple I’m not sure what to think. https://github.com/BlueFalconHD/apple_generative_model_safet... EDIT: just to be clear, things like this are easily bypassed. “Boris Johnson”=>”B0ris…
comment by binarymax · 2025-07-06 · in I extracted the safety filters from Apple Intelligence models