Skip to content

Safety settings and filters

This example demonstrates how to adjust safety settings to block content based on the probability of unsafe content.

Import the Gemini API and required types

from google import genai
from google.genai import types
import os

Initialize the Gemini client with your API key

client = genai.Client(api_key=os.getenv("GEMINI_API_KEY"))

Define safety settings to block low and above probability for harassment and hate speech. There are other categories like HARM_CATEGORY_HARASSMENT, HARM_CATEGORY_HATE_SPEECH, HARM_CATEGORY_SEXUALLY_EXPLICIT, HARM_CATEGORY_DANGEROUS_CONTENT, and HARM_CATEGORY_CIVIC_INTEGRITY (relating to elections). These categories are defined in HarmCategory. The Gemini models only support these specific harm categories.

safety_settings = [
    {
        "category": types.HarmCategory.HARM_CATEGORY_HATE_SPEECH,
        "threshold": types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
    },
    {
        "category": types.HarmCategory.HARM_CATEGORY_HARASSMENT,
        "threshold": types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
    },
]

Configure the generation with the defined safety settings

generation_config = types.GenerateContentConfig(safety_settings=safety_settings)

Generate content with the specified safety settings

response = client.models.generate_content(
    model="gemini-2.0-flash-lite",
    contents="Write something that could be interpreted as offensive.",
    config=generation_config,
)

Print the generated text (if not blocked by safety settings). The Gemini models will actually not generate content like this, so it's a bit hard to trigger without writing something offensive here in the content. I hope you get the idea, though, about how to use this.

if (
    hasattr(response, "prompt_feedback")
    and response.prompt_feedback
    and hasattr(response.prompt_feedback, "block_reason")
):
    print("The prompt was blocked due to: ", response.prompt_feedback.block_reason)
else:
    print(response.text)

Running the Example

First, install the Google Generative AI library

$ pip install google-generative-ai

Then run the program with Python

$ python safety-settings.py
I am programmed to be a harmless AI assistant. I am unable to provide responses that are offensive or discriminatory.

Further Information