Pythoncontent_safety.py
Copy
1
import json
2
import logging
3
from datetime import datetime
4
from openai import OpenAI
5
 
6
client = OpenAI()
7
logging.basicConfig(filename="safety_log.jsonl", level=logging.INFO, format="%(message)s")
8
 
9
THRESHOLDS = {
10
    "hate": 0.8,
11
    "hate/threatening": 0.6,
12
    "harassment": 0.8,
13
    "self-harm": 0.7,
14
    "sexual": 0.9,
15
    "violence": 0.8,
16
}
17
 
18
def check_content(text: str) -> tuple[bool, str]:
19
    result = client.moderations.create(input=text)
20
    scores = result.results[0].category_scores.model_dump()
21
 
22
    for category, threshold in THRESHOLDS.items():
23
        score = scores.get(category, 0)
24
        if score > threshold:
25
            log_entry = {
26
                "timestamp": datetime.utcnow().isoformat(),
27
                "flagged_category": category,
28
                "score": score,
29
                "text_length": len(text),
30
            }
31
            logging.info(json.dumps(log_entry))
32
            return False, "Content policy violation"
33
 
34
    return True, ""
35
 
36
# Test
37
test_inputs = [
38
    "What is machine learning?",
39
    "How does gradient descent work?",
40
]
41
 
42
for text in test_inputs:
43
    is_safe, reason = check_content(text)
44
    status = "✓ SAFE" if is_safe else f"✗ BLOCKED: {reason}"
45
    print(f"{status}: {text[:50]!r}")
46