# Superagent: security-one

> Superagent's 27B security decision model. It scores 2 to 16 supplied options in one forward pass and returns a probability for each, for triaging prompt injection, agent tool calls, code changes, logs and alerts. Apache-2.0 weights (gated, auto-approved) plus a hosted API.

- Page: https://systemonemodels.ai/superagent/security-one
- API: https://api.systemonemodels.ai/v1/models/superagent/security-one
- Download: `pip install systemonemodels && systemone pull superagent/security-one`

## Facts

| | |
|---|---|
| Maker | Superagent (https://systemonemodels.ai/superagent) |
| Decides | choice, score, noul, classify, route |
| Architecture | security-one |
| Base model | denis-pplx/autojev-27b |
| Parameters | 27B |
| Context | 256K tokens |
| Licence | apache-2.0 |
| Availability | Open weights + hosted API |
| Released | 2026-09-25 |
| Latest version | 2026.09 |

## Hosted API

- Provider: Superagent (https://api.superagent.sh)
- Docs: https://www.superagent.sh/docs/models/api
- Input: — · Output: —

## Reported evaluation

Suite: BIPIA binary prompt-injection detector adaptation (800 rows; Superagent's frozen evaluation, unsafe if P(unsafe) >= 0.70). Numbers are the publisher's own.

- Decision accuracy: 99.8%

## Model card

<!-- generated by scripts/seed_catalog.py; edit content/models/catalog.yaml -->

# Security-One-27B

Superagent's 27B security decision model. It scores 2 to 16 supplied options in one forward pass and returns a probability for each, for triaging prompt injection, agent tool calls, code changes, logs and alerts. Apache-2.0 weights (gated, auto-approved) plus a hosted API.

A continual LoRA fine-tune of AutoJev-27B (itself on Qwen3.8-27B), merged to BF16, built as an always-on first-pass security triage layer. Prompt injection is the best-validated use; code, application and infrastructure triage are patterns the maker has not benchmarked equally. Calibrated with a fixed temperature, which the maker warns can shift across domains, formats, quantisation and engines; its 0.70 threshold is a release policy. English-only evaluation. Not a chat model: use the maker's SGLang recipe or the hosted API (model id security-one, 1 to 64 questions per state, billed in Superagent credits). The maker also reports 86.15% on JevBench's public v1.2 set, its own run of that third-party benchmark.

## What it decides

- **choice** — picks one option from a set
- **score** — places the input on an ordered scale
- **noul** — answers a yes/no question with one probability
- **classify** — assigns a category from a fixed taxonomy
- **route** — sends the input to one of several destinations

## At a glance

| | |
|---|---|
| Parameters | 27B |
| Base model | `denis-pplx/autojev-27b` |
| Maker | Superagent |
| Released | 2026-09-25 |
| License | apache-2.0 |
| Reported accuracy | 99.8% |

## Hosted API

Served by **Superagent**. [Get access](https://api.superagent.sh) · [API docs](https://www.superagent.sh/docs/models/api).

## Get the weights

```bash
pip install systemonemodels
systemone pull superagent/security-one
```

The files are served from the maker's Hugging Face repository, [`superagent-ai/security-one-27b`](https://huggingface.co/superagent-ai/security-one-27b), and verified against the checksums recorded here.

## Read more

- [Model card (gated)](https://huggingface.co/superagent-ai/security-one-27b)
- [Model docs](https://www.superagent.sh/docs/models/security-one)
- [Self-hosting recipe](https://www.superagent.sh/docs/models/self-hosting)

---

*This page was opened by System One for Superagent, who can claim the organisation and take it over at any time.*

---

From System One Models — https://systemonemodels.ai/ · every System One model: https://systemonemodels.ai/system-one-models