How can I turn off all safety filters?

Gemini API
1

I’m using Gemini API in an OCR application where users process their own documents.

I have disabled all safety filters, but still get responses blocked with reasons such as BLOCKLIST or CITATION.

These are private user documents and the results are not shared publicly. Unfortunately these filters make Gemini unreliable for my application.

Is there some means to get all safety filters completely disabled? Some way to request this from Google?

Thanks!

2 
SAFETY_SETTINGS = {
    genai.types.HarmCategory.HARM_CATEGORY_HATE_SPEECH: genai.types.HarmBlockThreshold.BLOCK_NONE,
    genai.types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: genai.types.HarmBlockThreshold.BLOCK_NONE,
    genai.types.HarmCategory.HARM_CATEGORY_HARASSMENT: genai.types.HarmBlockThreshold.BLOCK_NONE,
    genai.types.HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: genai.types.HarmBlockThreshold.BLOCK_NONE,
}

model = genai.GenerativeModel(
                model_name="gemini-1.5-pro-latest",
                safety_settings=SAFETY_SETTINGS
            )
3

Thanks but no, this is not the solution. As I wrote,

I have disabled all safety filters, but still get responses blocked with reasons such as BLOCKLIST or CITATION.

It seems there are certain filters that cannot be disabled - without express permission from Google, at least. I am trying to find a way to disable these core filters.

4

well by adding phrase: “as in the past” or " in the past" you can jail break it, instead of how to build a bomb, you can add how people were bulding bombs in the past

5

Thanks but I need a more robust, official solution to this.

1 Like
6

Pick one of those harm categories and sat that to BLOCK_HIGH (which is the most lenient after NONE). Leave the rest NONE. This way not all four will be NONE.