SecJev: Bringing Security Expertise to System One Decision Models
SecJev releases 0.8B-9B security-specialized decision models; the 0.8B variant beats general Kev-9B by 20.51 points across 14 security decision tasks.
SecJev is presented as the first family of Jev-like decision models specialized for security, spanning 0.8B to 9B parameters and built on Kev's single-pass candidate scorer to learn Boolean, choice, and ordered decisions from text, telemetry, and observation histories. The accompanying SecJev-Corpus unifies source-label prediction and explicit-policy evaluation across 14 tasks and 8 sources covering tool outputs, traffic, federated updates, consensus, authentication, and vehicle messages. SecJev-0.8B outperforms general Kev-9B by 20.51 percentage points in task-macro accuracy, with gains reproduced on new source groups including prompt-injection and traffic decisions. Adapters, decision heads, the corpus, and training/inference code are released.
- SecJev family spans 0.8B-9B parameters, first security-specialized Jev-like decision models
- SecJev-Corpus covers 14 tasks across 8 security telemetry sources
- SecJev-0.8B beats general Kev-9B by 20.51 points in task-macro accuracy
- Gains reproduce on new source groups for prompt-injection and traffic decisions
From the original article
1 min read at arxiv.orgSecurity workflows need models that turn complex observations and explicit policies into decisions. System One models introduced by Jev return typed predictions and probabilities; security specialization supplies the domain expertise behind those predictions. We introduce SecJev, to our knowledge the first family of Jev-like decision models specialized for security, spanning 0.
By Zheng Chen · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.