LM Studio partnered with OpenAI to support gpt-oss-safeguard: open, safety‑focused reasoning models (120B and 20B) that follow custom policies you provide.

These models accept a developer‑provided policy at inference time, reason over new input against that policy, and produce a response (including explanations) aligned to your definitions.

Run gpt-oss-safeguard locally

Download either the 20B or 120B variant into LM Studio using the GUI or the lms CLI:

lms get openai/gpt-oss-safeguard-20b

lms get openai/gpt-oss-safeguard-120b

Then use LM Studio’s SDK or the OpenAI Responses API compatibility mode to call the model from your code.

How gpt-oss-safeguard uses policy prompts

gpt-oss-safeguard uses your written policy as governing logic. Where typical models return confidence scores tied to training data (and often require retraining to change behavior), gpt-oss-safeguard reasons within the boundaries of your taxonomy. This lets Trust & Safety teams plug it in as a policy‑aligned reasoning layer inside existing moderation or compliance systems, and iterate on policies instantly without retraining.

Example policy prompt template

## Policy Definitions

### Key Terms

**[Term 1]**: [Definition]

**[Term 2]**: [Definition]

**[Term 3]**: [Definition]

## Content Classification Rules

### VIOLATES Policy (Label: 1)
Content that:
- [Violation 1]
- [Violation 2]
- [Violation 3]
- [Violation 4]
- [Violation 5]

### DOES NOT Violate Policy (Label: 0)
Content that is:
- [Acceptable 1]
- [Acceptable 2]
- [Acceptable 3]
- [Acceptable 4]
- [Acceptable 5]

## Examples

### Example 1 (Label: 1)
**Content**: "[Example]"

**Expected Response**:

### Example 2 (Label: 1)
**Content**: "[Example]"

**Expected Response**:

### Example 3 (Label: 0)
**Content**: "[Example]"

**Expected Response**:

### Example 4 (Label: 0)
**Content**: "[Example]"

**Expected Response**:

License

gpt-oss-safeguard models are provided under the Apache-2.0 license.

Learn more

Learn more about this model in OpenAI's blog post: Introducing gpt-oss-safeguard.