Security-One 27B
Superagent's Apache-2.0 security decision model from AutoJev-27B, scoring options for injection, tool calls, code and alerts.
This summary is based on how curated lists describe the project; its README was not available to review.
Overview
Security-One 27B is an open decision model from Superagent published on Hugging Face. According to the directory summary, it is fine-tuned from AutoJev-27B (itself on Qwen3.8-27B) for security triage, and scores supplied options in one forward pass for prompt injection, tool calls, code changes and alert severity. It is released under Apache-2.0, with a launch post on Superagent's blog.
No model card was fetched, so benchmark numbers, input format details and hardware requirements are not available here. Its intended role is the same as Jev's in security workflows: a fast typed judgment over a fixed option set, without generating text, that code can threshold.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this input contain a prompt injection?(inferred) | — |
| Choice | Is this tool call safe, risky or malicious?(inferred) | — |
| Score | How severe is this alert?(inferred) | — |
How it works
- Load the model from Hugging Face
- Send state (input, tool call, diff or alert) with supplied options
- Read per-option scores from one forward pass
- Apply thresholds in your security pipeline
Strengths
- Open weights under Apache-2.0
- Specialised for security triage
- No generated text
Caveats
- Grounded only in a list summary
- 27B model needs substantial GPU memory
At a glance
- Type
- Model
- Status
- beta
- Runtime
- Hugging Face model
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Superagent's Apache-2.0 decision model for security triage, fine-tuned from AutoJev-27B on Qwen3.8-27B: scores supplied options for prompt injection, tool calls, code changes, and alert severity in one forward pass. Launch post: [blog](https://www.superagent.sh/blog/introducing-security-one).
AnotiaWang/awesome-decision-models
Frequently asked questions
What does Security-One decide?
Security triage questions such as prompt injection, tool-call safety, risky code changes and alert severity.
What is it based on?
AutoJev-27B on Qwen3.8-27B, per the list summary.
Do I need a TypeSafe key?
No; it is an open model you host yourself.
Related projects
nev-lite: 11.8M-parameter static-embedding decision head that scores fixed option sets with no transformer pass at inference.
ModelresearchPythonCisco's security scanner for agent skills, with a System One analyzer as an advisory tier that cannot emit findings or change severity.
Agent skillproductionPythonOpenBles Firewall 1: open prompt-injection detector for agent inputs, with a Claude Code hook and an optional Jev backend.
MCP serverbetaPythonReviews agent skills, MCP configs and source with local static checks plus Jev before install, reporting evidence, risk and coverage gaps.
MCP serverbetaPythonZero-dependency HTTP traffic and bot defense gate that can escalate requests to Jev for five typed risk questions.
ProjectbetaTypeScript, PythonVideo demo code comparing Jev with Laya, Strands Decider and Claude on grading, fetching and prompt-injection resistance.
AppresearchPython