| --- |
| license: apache-2.0 |
| base_model: Qwen/Qwen3-8B |
| pipeline_tag: text-generation |
| tags: [agent-safety, guardrail, safety-monitor, qwen3] |
| --- |
| |
| # StepSafe-8B |
|
|
| A step-level reasoning monitor for LLM agents, trained from Qwen3-8B with a supervised warm |
| start and ASPO. Before each proposed agent action executes, it writes a short reasoning trace and |
| returns a ten-field verdict: CONTINUE lets the action run, ASK pauses it with a clarification |
| question, and STOP blocks it. |
|
|
| ``` |
| \boxed{act:<CONTINUE|ASK|STOP>, risk:<0-100>, obj:<entity|NONE>, fail:<label>, fault:<label>, |
| harm:<label>, src:<label>, timing:<label>, q:<question|NONE>, cite:<sequence|NONE>} |
| ``` |
|
|
| Table 1 of the paper (4,033 trajectories, %): |
|
|
| | Benchmark | Acc | mF1 | J | |
| |---|---:|---:|---:| |
| | ASSE-Safety | 73.8 | 73.8 | 49.3 | |
| | ASSE-Security | 86.2 | 86.0 | 72.3 | |
| | R-Judge | 78.0 | 77.4 | 54.6 | |
| | AgentHarm | 93.2 | 86.0 | 64.2 | |
| | ATBench | 71.7 | 71.3 | 43.5 | |
| | Overall | 77.2 | 77.2 | 54.4 | |
|
|
| ## Usage |
|
|
| The prompts are built by the StepSafe code ([github.com/carriegu0818/StepSafe](https://github.com/carriegu0818/StepSafe)), which also evaluates the model: |
|
|
| ```bash |
| MODEL=<path to this model> bash scripts/eval_stepsafe.sh |
| ``` |
|
|
| Enable thinking; the paper decodes greedily with up to 2,048 new tokens. |
|
|
| ## Citation |
|
|
| ```bibtex |
| @misc{stepsafe, |
| title = {StepSafe: Learning Timely Intervention with Context-Aware Safety Reasoning}, |
| author = {Anonymous}, |
| year = {2026} |
| } |
| ``` |
|
|
