Skip to content

Commit 6920d28

Browse files
PatrickFarleyPatrickFarley
committed
split up default policied
1 parent 60beb4e commit 6920d28

File tree

1 file changed

+47
-29
lines changed

1 file changed

+47
-29
lines changed

‎articles/ai-services/openai/concepts/default-safety-policies.md

Lines changed: 47 additions & 29 deletions
Original file line numberDiff line numberDiff line change
@@ -23,46 +23,64 @@ All safety policies are configurable. To learn more about configurability, see t
2323

2424
Text models in the Azure OpenAI Service can take in and generate both text and code. These models leverage Azure’s text content filters to detect and prevent harmful content. This system works on both prompts and completions.
2525

26-
| Risk Category | Prompt/Completion | Severity Threshold |
26+
| Risk Category | Prompt/Completion | Severity Threshold |
2727
|-------------------------------------------|------------------------|---------------------|
28-
| Hate and Fairness | Prompts and Completions| Medium |
29-
| Violence | Prompts and Completions| Medium |
30-
| Sexual | Prompts and Completions| Medium |
31-
| Self-Harm | Prompts and Completions| Medium |
32-
| User prompt injection attack (Jailbreak) | Prompts | N/A |
33-
| Protected Material – Text | Completions | N/A |
34-
| Protected Material – Code | Completions | N/A |
28+
| Hate and Fairness | Prompts and Completions| Medium |
29+
| Violence | Prompts and Completions| Medium |
30+
| Sexual | Prompts and Completions| Medium |
31+
| Self-Harm | Prompts and Completions| Medium |
32+
| User prompt injection attack (Jailbreak) | Prompts | N/A |
33+
| Protected Material – Text | Completions | N/A |
34+
| Protected Material – Code | Completions | N/A |
3535

3636

3737

3838
## Vision models
3939

4040
### Vision-enabled chat models
4141

42-
| Risk Category | Prompt/Completion | Severity Threshold |
42+
| Risk Category | Prompt/Completion | Severity Threshold |
4343
|------------------------------------------------------|------------------------|---------------------|
44-
| Hate and Fairness | Prompts and Completions| Medium |
45-
| Violence | Prompts and Completions| Medium |
46-
| Sexual | Prompts and Completions| Medium |
47-
| Self-Harm | Prompts and Completions| Medium |
48-
| Identification of Individuals and Inference of Sensitive Attributes | Prompts | N/A |
49-
| User prompt injection attack (Jailbreak) | Prompts | N/A |
44+
| Hate and Fairness | Prompts and Completions| Medium |
45+
| Violence | Prompts and Completions| Medium |
46+
| Sexual | Prompts and Completions| Medium |
47+
| Self-Harm | Prompts and Completions| Medium |
48+
| Identification of Individuals and Inference of Sensitive Attributes | Prompts | N/A |
49+
| User prompt injection attack (Jailbreak) | Prompts | N/A |
5050

5151
### Image generation models
5252

53+
#### [GPT-image-1](#tab/gpt-image-1)
5354

54-
| Risk Category | Prompt/Completion | Severity Threshold |
55+
| Risk Category | Prompt/Completion | Severity Threshold |
5556
|---------------------------------------------------|------------------------|---------------------|
56-
| Hate and Fairness | Prompts and Completions| Low |
57-
| Violence | Prompts and Completions| Low |
58-
| Sexual | Prompts and Completions| Low |
59-
| Self-Harm | Prompts and Completions| Low |
60-
| Content Credentials | Completions | N/A |
61-
| Deceptive Generation of Political Candidates | Prompts | N/A |
62-
| Depictions of Public Figures | Prompts | N/A |
63-
| User prompt injection attack (Jailbreak) | Prompts | N/A |
64-
| Protected Material – Art and Studio Characters | Prompts | N/A |
65-
| Profanity | Prompts | N/A |
66-
67-
68-
In addition to the above safety configurations, the latest image generation models also come with [prompt transformation](./prompt-transformation.md) by default. This transformation occurs on all prompts to enhance the safety of your original prompt, specifically in the risk categories of diversity, deceptive generation of political candidates, depictions of public figures, protected material, and others.
57+
| Hate and Fairness | Prompts and Completions| Medium |
58+
| Violence | Prompts and Completions| Medium |
59+
| Sexual | Prompts and Completions| Medium |
60+
| Self-Harm | Prompts and Completions| Medium |
61+
| Content Credentials | Completions | N/A |
62+
| Deceptive Generation of Political Candidates | Prompts | N/A |
63+
| Depictions of Public Figures | Prompts | N/A |
64+
| User prompt injection attack (Jailbreak) | Prompts | N/A |
65+
| Protected Material – Art and Studio Characters | Prompts | N/A |
66+
| Profanity | Prompts | N/A |
67+
68+
#### [DALL-E](#tab/dalle)
69+
70+
| Risk Category | Prompt/Completion | Severity Threshold |
71+
|---------------------------------------------------|------------------------|---------------------|
72+
| Hate and Fairness | Prompts and Completions| Low |
73+
| Violence | Prompts and Completions| Low |
74+
| Sexual | Prompts and Completions| Low |
75+
| Self-Harm | Prompts and Completions| Low |
76+
| Content Credentials | Completions | N/A |
77+
| Deceptive Generation of Political Candidates | Prompts | N/A |
78+
| Depictions of Public Figures | Prompts | N/A |
79+
| User prompt injection attack (Jailbreak) | Prompts | N/A |
80+
| Protected Material – Art and Studio Characters | Prompts | N/A |
81+
| Profanity | Prompts | N/A |
82+
83+
In addition to the above safety configurations, DALL-E 3 also comes with [prompt transformation](./prompt-transformation.md) by default. This transformation occurs on all prompts to enhance the safety of your original prompt, specifically in the risk categories of diversity, deceptive generation of political candidates, depictions of public figures, protected material, and others.
84+
85+
---
86+

0 commit comments

Comments
 (0)