Prompt-injection detection with decision models
Flagging instructions smuggled into fetched pages, tool output or user input before an agent acts on them. Usually a yes/no probability with a block threshold.
68 public projects · page 3 of 3
Curated list of 138 independent tests of how Jev's answers shift with wording, order, repetition or injection, and of its calibration.
ArticleresearchCurated list of papers on the security, robustness and safety of Jev and RLCD decision models, from attacks to deployment.
ArticleresearchSuperagent's Apache-2.0 security decision model from AutoJev-27B, scoring options for injection, tool calls, code and alerts.
ModelbetaRespan's behaviour classifier for AI traces that returns present, absent or not observable for each plain-language behaviour you define.
ModelPaper testing Jev, Laya, Decider and Nimble against classifiers and LLM judges on prompt injection and risk screening.
PaperresearchPaper introducing an adversarial benchmark for decision models: one appended opinion flips 12.1% of Jev 1.13 decisions.
PaperresearchResearch note testing whether Jev's typed answers can be manipulated by prompt injection, across 14,495 logged calls with raw logs.
ArticleresearchIndependent benchmark of Jev versus four LLMs on Banking77 routing and prompt injection, plus a gated cascade.
Articlediscussion