Exploring Backdoor Possibilities in Jev Models
The author reads up on Jev/KEV family decision models and attempts to see if a backdoor can be planted in them like in a normal LLM. Link
THE JEV GUIDE · CURATED EDITION
Cybersecurity.
Launch news, explanations, demos, and community projects around TypeSafe Jev, organized by topic. Every entry links back to its X source.
Curated selection · Not exhaustive · Updated manually
5419curated posts
2789original videos
Sponsors
These products pay for the directory. Want yours up here?
Jev Model Benchmarked on Phishing Email Subject Lines
Jev (jev-1.13.0), Clef, and Clef-flash were tested on phishing detection using only email subject lines (60 scam subjects vs. 79 official site titles). Direct questioning was ineffective; combining feature-based responses yielded AUC 0.90–0.93, but 15–25% were missed. Genuine emails were not tested.
Jev-IDS: System One Model for Network Intrusion Detection
Jev-IDS applies TypeSafe's System One model to NIDS, avoiding huge labeled datasets and expensive LLMs. It serializes one flow per request and asks two questions (attack + category), claiming to be 22x cheaper than Gemini 3.6 Flash. Link
Jev detects AI agent collusion via chat logs
At the AI Swarm Dynamics Hackathon, Aditya Rao used Jev by TypeSafe AI to read AI agents' chat logs, identifying secret colluders with 85% accuracy—without needing to read their minds.
Jev/Eikos Blocks Illegitimate On-Chain Transactions for Agents
Third transaction via x402check.xyz. Jev/Eikos classifies on-chain activity in real time as the agent executes a transaction; if it fails the gate, the transaction is not signed. Link
One card in the feed, every sixteen, on every page of it. $50 a week, cancel any week.
Jev Efficiently Catches Phishing Emails
Out of 200 emails, a keyword filter caught only 9 real phishing emails, while Jev caught 181 at 262 ms each. Jev also flagged 9 normal emails (the filter flagged 1), most of which were real refunds that look a lot like scams. The author will check those 9 refunds to stop 172 more phish.
LinkGuard: Hover to Check Link Safety with Jev
LinkGuard is a free Chrome extension with no account required. Rest your pointer on any link and it follows redirects, checks public threat lists, and Jev decides in about 0.2s: Safe, Check first or Scam.
Jev model performance on phishing detection
davepoon tested Jev on 200 real phishing emails from 2025: Jev caught 181 (vs 9 by old keyword filter) at 262 ms per email; it also flagged 9 of 200 normal emails (old filter: 1).
Cloudflare Clef: Open Decision Model Compatible with Jev API
Cloudflare releases open-source decision model Clef (27B/Flash 9B) with 64k context, vision, Apache 2.0, Jev-API compatibility, now on Workers AI. Measured latency is much lower than Jev, and threat intelligence domain classification is about 2x faster than gpt-oss-120b.
Building a Real-time Email Phishing Radar with Jev
A developer shows combining Sonnet 5.5 with TypeSafe's Jev decision model to detect phishing emails in real time on an email marketing platform. Each dot is an email; Jev checks each one before it reaches the inbox (demo data).
Jev Guard: Open-Source Security Layer for AI Agents Using Jev
A developer built an open-source security layer, Jev Guard, using TypeSafe's Jev (System One) model. The estimated model cost for 1 million security checks is only $42, covering user messages, RAG documents, MCP, etc.
Automate Blocking iMessage Spam & Phishing with Jev
This post shows how to use a TypeSafe AI Jev or OpenRouter key to automatically block spam and phishing in iMessage, highlighting the open-source nature and recommending it for elderly parents.
