About

An independent bilingual platform for current AI security news, research, vulnerabilities and weekly context

How fy(ai)-security works

Technical content flow from source intake through Ollama and quality gates to editorial approval

  1. 01
    IntakeApproved sources

    Fixed adapters collect vendor feeds, security media, arXiv, NVD, CISA and advisories as inherently untrusted input

  2. 02
    IntakeExtraction & normalization

    Python collectors extract visible content, normalize URLs, timestamps and metadata, and reject disallowed destinations

  3. 03
    PipelineFiltering & clustering

    Time windows and relevance rules limit candidates. Canonical URLs, stable identifiers and source clusters reduce duplicates

  4. 04
    AI PipelineOllama analysis

    An open-source model assesses relevance, classifies the content type and produces a bilingual structured draft

  5. 05
    Quality GateGrounding & validation

    Pydantic schemas, deterministic rules and a separate grounding pass verify required fields, source support, date logic and uncertainty

  6. 06
    BackendProtected import API

    An authenticated, idempotent Next.js API accepts validated batches only. Zod, size limits, rate limits and markup checks protect the boundary

  7. 07
    EditorialHuman editorial review

    Every draft is reviewed by a person before publication. Content can be revised, approved or rejected

  8. 08
    PublicationContent store & frontend

    The service layer stores versions, sources and state in an SQL database. The public frontend reads approved records only

Portrait of Sebastian Grünewald

About Me

Hello fellow AI Security nerds, enthusiasts and researchers — welcome!

My name is Sebastian Grünewald and I lead an AI Security team in consulting and pursue AI security research out of genuine passion. My background started in secure software development, later moved into senior security architecture, and eventually led me to research work at UC Berkeley, where my interest in AI security became a central focus.

The idea for fy(a)i-security came from a simple observation: it is becoming increasingly difficult to keep up with the pace of AI, security research, new vulnerabilities and emerging attack techniques. At the same time, I felt there was no focused platform dedicated specifically to this niche. With fy(a)i-security, I want to make relevant AI security developments easier to follow — and also test, in practice, how far modern web applications can be built with the help of coding assistants like Codex.