azure-ai-contentsafety-py
用 Python 版 Content Safety SDK 偵測文字圖片的有害內容
中文翻譯
---
name: azure-ai-contentsafety-py
description: |
Azure AI Content Safety SDK（Python 版）。用於偵測文字和圖片中的有害內容，提供多級嚴重程度分類。
Triggers: "azure-ai-contentsafety", "ContentSafetyClient", "content moderation", "harmful content", "text analysis", "image analysis"。
license: MIT
metadata:
author: Microsoft
version: "1.0.0"
package: azure-ai-contentsafety
---
# Azure AI Content Safety SDK（Python）
在應用程式裡偵測使用者產生和 AI 產生的有害內容。
## 安裝
```bash
pip install azure-ai-contentsafety
```
## 環境變數
```bash
CONTENT_SAFETY_ENDPOINT=https://<resource>.cognitiveservices.azure.com # 所有認證方式都必填
AZURE_TOKEN_CREDENTIALS=prod # 只有在正式環境用 DefaultAzureCredential 時才需要
CONTENT_SAFETY_KEY=<your-api-key> # 只有用下面舊的 API key 認證路徑時才需要
```
## 認證與生命週期
> **🔑 下面每一段範例程式都適用兩條規則：**
>
> 1. **優先用 `DefaultAzureCredential`。** 它在本機（Azure CLI / VS Code / Developer CLI）和 Azure 上（managed identity、workload identity）都能跑，程式碼不用改。避免用 connection string、帳號金鑰或 API key，它們會繞過 Entra 的稽核和金鑰輪替。
> - 本機開發：直接用 `DefaultAzureCredential` 就好。
> - 正式環境：設定 `AZURE_TOKEN_CREDENTIALS=prod`（或 `AZURE_TOKEN_CREDENTIALS=<specific_credential>`），把 credential chain 限縮在適合正式環境的憑證。
> 2. **每個 client 都要包在 context manager 裡**，這樣 HTTP transport、socket 和 token cache 才會確實被釋放：
> - 同步：`with <Client>(...) as client:`
> - 非同步：`async with <Client>(...) as client:` **以及** `async with DefaultAzureCredential() as credential:`（來自 `azure.identity.aio`）
>
> 範例片段可能會省略這些設定，但正式環境的程式碼一定要遵守這兩條規則。
```python
import os
from azure.identity import DefaultAzureCredential, ManagedIdentityCredential
from azure.ai.contentsafety import ContentSafetyClient
from azure.ai.contentsafety.models import AnalyzeTextOptions
# 本機開發：用 DefaultAzureCredential。正式環境：設定 AZURE_TOKEN_CREDENTIALS=prod 或 AZURE_TOKEN_CREDENTIALS=<specific_credential>
credential = DefaultAzureCredential(require_envvar=True)
# 或在正式環境直接使用指定的 credential：
# See https://learn.microsoft.com/python/api/overview/azure/identity-readme?view=azure-python#credential-classes
# credential = ManagedIdentityCredential()
with ContentSafetyClient(
endpoint=os.environ["CONTENT_SAFETY_ENDPOINT"],
credential=credential,
) as client:
response = client.analyze_text(AnalyzeTextOptions(text="Hello, world!"))
```
### 舊做法：API Key（既有使用金鑰的部署）
新的程式碼應該用上面的 `DefaultAzureCredential`。只有在既有的金鑰部署還沒遷移到 Entra ID 時才用 `AzureKeyCredential`，例如還在完成 Entra 導入的受規範環境。
```python
import os
from azure.core.credentials import AzureKeyCredential
from azure.ai.contentsafety import ContentSafetyClient
from azure.ai.contentsafety.models import AnalyzeTextOptions
with ContentSafetyClient(
endpoint=os.environ["CONTENT_SAFETY_ENDPOINT"],
credential=AzureKeyCredential(os.environ["CONTENT_SAFETY_KEY"]),
) as client:
response = client.analyze_text(AnalyzeTextOptions(text="Hello, world!"))
```
如果你也需要用金鑰管理 blocklist，`BlocklistClient` 接受同樣的 `AzureKeyCredential`。
## 分析文字
```python
from azure.ai.contentsafety import ContentSafetyClient
from azure.ai.contentsafety.models import AnalyzeTextOptions, TextCategory
from azure.identity import DefaultAzureCredential
with ContentSafetyClient(endpoint, DefaultAzureCredential()) as client:
request = AnalyzeTextOptions(text="Your text content to analyze")
response = client.analyze_text(request)
# 檢查每個類別
for category in [TextCategory.HATE, TextCategory.SELF_HARM,
TextCategory.SEXUAL, TextCategory.VIOLENCE]:
result = next((r for r in response.categories_analysis
if r.category == category), None)
if result:
print(f"{category}: severity {result.severity}")
```
## 分析圖片
```python
from azure.ai.contentsafety import ContentSafetyClient
from azure.ai.contentsafety.models import AnalyzeImageOptions, ImageData
from azure.identity import DefaultAzureCredential
import base64
with ContentSafetyClient(endpoint, DefaultAzureCredential()) as client:
# 從檔案讀取
with open("image.jpg", "rb") as f:
image_data = base64.b64encode(f.read()).decode("utf-8")
request = AnalyzeImageOptions(
image=ImageData(content=image_data)
)
response = client.analyze_image(request)
for result in response.categories_analysis:
print(f"{result.category}: severity {result.severity}")
```
### 網址上的圖片
```python
from azure.ai.contentsafety.models import AnalyzeImageOptions, ImageData
request = AnalyzeImageOptions(
image=ImageData(blob_url="https://example.com/image.jpg")
)
response = client.analyze_image(request)
```
## 文字 Blocklist 管理
### 建立 Blocklist
```python
from azure.ai.contentsafety import BlocklistClient
from azure.ai.contentsafety.models import TextBlocklist
from azure.identity import DefaultAzureCredential
with BlocklistClient(endpoint, DefaultAzureCredential()) as blocklist_client:
blocklist = TextBlocklist(
blocklist_name="my-blocklist",
description="Custom terms to block"
)
result = blocklist_client.create_or_update_text_blocklist(
blocklist_name="my-blocklist",
options=blocklist
)
```
### 新增封鎖項目
```python
from azure.ai.contentsafety.models import AddOrUpdateTextBlocklistItemsOptions, TextBlocklistItem
items = AddOrUpdateTextBlocklistItemsOptions(
blocklist_items=[
TextBlocklistItem(text="blocked-term-1"),
TextBlocklistItem(text="blocked-term-2")
]
)
result = blocklist_client.add_or_update_blocklist_items(
blocklist_name="my-blocklist",
options=items
)
```
### 搭配 Blocklist 分析
```python
from azure.ai.contentsafety.models import AnalyzeTextOptions
request = AnalyzeTextOptions(
text="Text containing blocked-term-1",
blocklist_names=["my-blocklist"],
halt_on_blocklist_hit=True
)
response = client.analyze_text(request)
if response.blocklists_match:
for match in response.blocklists_match:
print(f"Blocked: {match.blocklist_item_text}")
```
## 嚴重程度
文字分析預設回傳 4 級嚴重程度（0、2、4、6）。要用 8 級（0-7）：
```python
from azure.ai.contentsafety.models import AnalyzeTextOptions, AnalyzeTextOutputType
request = AnalyzeTextOptions(
text="Your text",
output_type=AnalyzeTextOutputType.EIGHT_SEVERITY_LEVELS
)
```
## 有害內容類別
| 類別 | 說明 |
|----------|-------------|
| `Hate` | 針對身分（種族、宗教、性別等）的攻擊 |
| `Sexual` | 性內容、關係、身體部位 |
| `Violence` | 肢體傷害、武器、受傷 |
| `SelfHarm` | 自傷、自殺、飲食失調 |
## 嚴重程度級距
| 級別 | 文字範圍 | 圖片範圍 | 意義 |
|-------|------------|-------------|---------|
| 0 | Safe | Safe | 沒有有害內容 |
| 2 | Low | Low | 輕微提及 |
| 4 | Medium | Medium | 中等程度的內容 |
| 6 | High | High | 嚴重的內容 |
## Client 類型
| Client | 用途 |
|--------|---------|
| `ContentSafetyClient` | 分析文字和圖片 |
| `BlocklistClient` | 管理自訂 blocklist |
## 最佳實務
1. **同步或非同步擇一，並保持一致。** 同一條呼叫路徑裡，不要混用 `azure.ai.contentsafety` 的同步 client 和 `azure.ai.contentsafety.aio` 的非同步 client。每個模組選一種模式。
2. **client 和非同步 credential 一律用 context manager。** 每個 client 都包在 `with ContentSafetyClient(...) as client:`（同步）或 `async with ContentSafetyClient(...) as client:`（非同步）裡。來自 `azure.identity.aio` 的非同步 `DefaultAzureCredential` 也要用 `async with credential:`，token 和 transport 才會被清掉。
3. 特定領域的詞彙**用 blocklist**
4. **設定嚴重程度門檻**，要符合你的使用情境
5. **處理多個類別**：內容可能在好幾方面都有害
6. 要立即拒絕時**用 halt_on_blocklist_hit**
7. **記錄分析結果**，供稽核和改進使用
8. 需要更細的控制時，**考慮 8 級嚴重程度模式**
9. AI 的輸出在顯示給使用者之前**先審核**
## 參考檔案
| 檔案 | 內容 |
|------|----------|
| [references/capabilities.md](references/capabilities.md) | 其他非主打的功能、各 operation group 的涵蓋範圍，以及正式環境檢查清單。 |
| [references/non-hero-scenarios.md](references/non-hero-scenarios.md) | 針對次要／進階情境的專門範例（非主打）。 |英文原文
---
name: azure-ai-contentsafety-py
description: |
Azure AI Content Safety SDK for Python. Use for detecting harmful content in text and images with multi-severity classification.
Triggers: "azure-ai-contentsafety", "ContentSafetyClient", "content moderation", "harmful content", "text analysis", "image analysis".
license: MIT
metadata:
author: Microsoft
version: "1.0.0"
package: azure-ai-contentsafety
---
# Azure AI Content Safety SDK for Python
Detect harmful user-generated and AI-generated content in applications.
## Installation
```bash
pip install azure-ai-contentsafety
```
## Environment Variables
```bash
CONTENT_SAFETY_ENDPOINT=https://<resource>.cognitiveservices.azure.com # Required for all auth methods
AZURE_TOKEN_CREDENTIALS=prod # Required only if DefaultAzureCredential is used in production
CONTENT_SAFETY_KEY=<your-api-key> # Only required for the legacy API-key auth path below
```
## Authentication & Lifecycle
> **🔑 Two rules apply to every code sample below:**
>
> 1. **Prefer `DefaultAzureCredential`.** It works locally (Azure CLI / VS Code / Developer CLI) and in Azure (managed identity, workload identity) with no code change. Avoid connection strings, account/API keys — they bypass Entra audit and rotation.
> - Local dev: `DefaultAzureCredential` works as-is.
> - Production: set `AZURE_TOKEN_CREDENTIALS=prod` (or `AZURE_TOKEN_CREDENTIALS=<specific_credential>`) to constrain the credential chain to production-safe credentials.
> 2. **Wrap every client in a context manager** so HTTP transports, sockets, and token caches are released deterministically:
> - Sync: `with <Client>(...) as client:`
> - Async: `async with <Client>(...) as client:` **and** `async with DefaultAzureCredential() as credential:` (from `azure.identity.aio`)
>
> Snippets may abbreviate this setup, but production code should always follow both rules.
```python
import os
from azure.identity import DefaultAzureCredential, ManagedIdentityCredential
from azure.ai.contentsafety import ContentSafetyClient
from azure.ai.contentsafety.models import AnalyzeTextOptions
# Local dev: DefaultAzureCredential. Production: set AZURE_TOKEN_CREDENTIALS=prod or AZURE_TOKEN_CREDENTIALS=<specific_credential>
credential = DefaultAzureCredential(require_envvar=True)
# Or use a specific credential directly in production:
# See https://learn.microsoft.com/python/api/overview/azure/identity-readme?view=azure-python#credential-classes
# credential = ManagedIdentityCredential()
with ContentSafetyClient(
endpoint=os.environ["CONTENT_SAFETY_ENDPOINT"],
credential=credential,
) as client:
response = client.analyze_text(AnalyzeTextOptions(text="Hello, world!"))
```
### Legacy: API Key (existing keyed deployments)
New code should use `DefaultAzureCredential` above. Use `AzureKeyCredential` only if you have an existing keyed deployment that hasn't been migrated to Entra ID yet — for example, regulated environments still completing their Entra rollout.
```python
import os
from azure.core.credentials import AzureKeyCredential
from azure.ai.contentsafety import ContentSafetyClient
from azure.ai.contentsafety.models import AnalyzeTextOptions
with ContentSafetyClient(
endpoint=os.environ["CONTENT_SAFETY_ENDPOINT"],
credential=AzureKeyCredential(os.environ["CONTENT_SAFETY_KEY"]),
) as client:
response = client.analyze_text(AnalyzeTextOptions(text="Hello, world!"))
```
The `BlocklistClient` accepts the same `AzureKeyCredential` if you also need to manage blocklists with a key.
## Analyze Text
```python
from azure.ai.contentsafety import ContentSafetyClient
from azure.ai.contentsafety.models import AnalyzeTextOptions, TextCategory
from azure.identity import DefaultAzureCredential
with ContentSafetyClient(endpoint, DefaultAzureCredential()) as client:
request = AnalyzeTextOptions(text="Your text content to analyze")
response = client.analyze_text(request)
# Check each category
for category in [TextCategory.HATE, TextCategory.SELF_HARM,
TextCategory.SEXUAL, TextCategory.VIOLENCE]:
result = next((r for r in response.categories_analysis
if r.category == category), None)
if result:
print(f"{category}: severity {result.severity}")
```
## Analyze Image
```python
from azure.ai.contentsafety import ContentSafetyClient
from azure.ai.contentsafety.models import AnalyzeImageOptions, ImageData
from azure.identity import DefaultAzureCredential
import base64
with ContentSafetyClient(endpoint, DefaultAzureCredential()) as client:
# From file
with open("image.jpg", "rb") as f:
image_data = base64.b64encode(f.read()).decode("utf-8")
request = AnalyzeImageOptions(
image=ImageData(content=image_data)
)
response = client.analyze_image(request)
for result in response.categories_analysis:
print(f"{result.category}: severity {result.severity}")
```
### Image from URL
```python
from azure.ai.contentsafety.models import AnalyzeImageOptions, ImageData
request = AnalyzeImageOptions(
image=ImageData(blob_url="https://example.com/image.jpg")
)
response = client.analyze_image(request)
```
## Text Blocklist Management
### Create Blocklist
```python
from azure.ai.contentsafety import BlocklistClient
from azure.ai.contentsafety.models import TextBlocklist
from azure.identity import DefaultAzureCredential
with BlocklistClient(endpoint, DefaultAzureCredential()) as blocklist_client:
blocklist = TextBlocklist(
blocklist_name="my-blocklist",
description="Custom terms to block"
)
result = blocklist_client.create_or_update_text_blocklist(
blocklist_name="my-blocklist",
options=blocklist
)
```
### Add Block Items
```python
from azure.ai.contentsafety.models import AddOrUpdateTextBlocklistItemsOptions, TextBlocklistItem
items = AddOrUpdateTextBlocklistItemsOptions(
blocklist_items=[
TextBlocklistItem(text="blocked-term-1"),
TextBlocklistItem(text="blocked-term-2")
]
)
result = blocklist_client.add_or_update_blocklist_items(
blocklist_name="my-blocklist",
options=items
)
```
### Analyze with Blocklist
```python
from azure.ai.contentsafety.models import AnalyzeTextOptions
request = AnalyzeTextOptions(
text="Text containing blocked-term-1",
blocklist_names=["my-blocklist"],
halt_on_blocklist_hit=True
)
response = client.analyze_text(request)
if response.blocklists_match:
for match in response.blocklists_match:
print(f"Blocked: {match.blocklist_item_text}")
```
## Severity Levels
Text analysis returns 4 severity levels (0, 2, 4, 6) by default. For 8 levels (0-7):
```python
from azure.ai.contentsafety.models import AnalyzeTextOptions, AnalyzeTextOutputType
request = AnalyzeTextOptions(
text="Your text",
output_type=AnalyzeTextOutputType.EIGHT_SEVERITY_LEVELS
)
```
## Harm Categories
| Category | Description |
|----------|-------------|
| `Hate` | Attacks based on identity (race, religion, gender, etc.) |
| `Sexual` | Sexual content, relationships, anatomy |
| `Violence` | Physical harm, weapons, injury |
| `SelfHarm` | Self-injury, suicide, eating disorders |
## Severity Scale
| Level | Text Range | Image Range | Meaning |
|-------|------------|-------------|---------|
| 0 | Safe | Safe | No harmful content |
| 2 | Low | Low | Mild references |
| 4 | Medium | Medium | Moderate content |
| 6 | High | High | Severe content |
## Client Types
| Client | Purpose |
|--------|---------|
| `ContentSafetyClient` | Analyze text and images |
| `BlocklistClient` | Manage custom blocklists |
## Best Practices
1. **Pick sync OR async and stay consistent.** Do not mix `azure.ai.contentsafety` sync clients with `azure.ai.contentsafety.aio` async clients in the same call path. Choose one mode per module.
2. **Always use context managers for clients and async credentials.** Wrap every client in `with ContentSafetyClient(...) as client:` (sync) or `async with ContentSafetyClient(...) as client:` (async). For async `DefaultAzureCredential` from `azure.identity.aio`, also use `async with credential:` so tokens and transports are cleaned up.
3. **Use blocklists** for domain-specific terms
4. **Set severity thresholds** appropriate for your use case
5. **Handle multiple categories** — content can be harmful in multiple ways
6. **Use halt_on_blocklist_hit** for immediate rejection
7. **Log analysis results** for audit and improvement
8. **Consider 8-severity mode** for finer-grained control
9. **Pre-moderate AI outputs** before showing to users
## Reference Files
| File | Contents |
|------|----------|
| [references/capabilities.md](references/capabilities.md) | Additional non-hero capabilities, operation-group coverage, and production checklists. |
| [references/non-hero-scenarios.md](references/non-hero-scenarios.md) | Dedicated non-hero examples for secondary/advanced scenarios. |