Jev Security Review Demos

Mapping connections across 1,010 X posts and grouping them into 22 suspected scam syndicates

Classifying page elements without filter lists, removing 16 ads, and caching the decisions for the next visit

Evaluating 1,968 skincare and wellness ads with 33 typed questions each, producing 64,944 answers in 36 seconds

Measuring choice probabilities across labeled ethical dilemmas involving age, race, gender, humans, and AI

Scoring the probability that a Tibo post will trigger a Codex quota reset

Classifying scam email with configurable parameters and thresholds in an open-source safety proof of concept

Reviewing a patch for semantic harm and flagging three test-masking changes in 343 milliseconds

Adding decision gates to agents that check SEO content, recruiting evidence, and produced videos before action

Checking a ChatGPT-drafted document end to end and finding explanations unsupported by the cited source

Detecting thousands of multilingual bad-word spellings and variants for real-time chat moderation

Scoring command operational security in Mythic in real time and identifying the expression type

Testing an AI request gate against eight attacks including fake admin claims, secret extraction, and harmless quoted threats

Classifying a URL as safe, suspicious, or scam from its address, anchor text, and page title in about 0.5 seconds

Classifying page content in real time and blocking material that violates a child's safety policy

Detecting personally identifiable information before text reaches another AI and comparing classification with semantic lookup tasks

Returning structured JSON decisions for phone-call intent and dog sounds entirely on edge devices

Judging coding-agent tool calls as allow, ask or block, with rules rejecting known threats in about 40 ms

Inspecting faces, fingers, text, numbers and poses to flag likely AI-generated photos

Moderating 100 comments in real time for $0.002171 with calibrated typed judgments

Moderating live text directly and routing local OCR or model descriptions for images and video

Flagging command risk as text is typed and evaluating 200 labeled commands

Scoring every proposed agent action on six safety factors before choosing act or hold for $0.001 per decision

Filtering citations from a poisoned RAG corpus to keep agents from citing low-quality material

Checking Supabase RLS policies against one rubric while comparing Jev, GPT and Gemini on leaks, speed and cost