home.social
  1. Everyone’s hyping #GTA6 – I’m still thinking about the Rockstar breach. 🎮

    Billion-dollar budget, top devs. The entry point? Social engineering & a single Slack account.

    The priciest firewall won’t save you if Zero Trust, IAM, and Least Privilege stop at internal tools. Hotwiring a car in Los Santos shouldn't be harder than stealing access rights. 😉

    IT review services: m-ko.de

    #ITSecurity #InfoSec #CyberSecurity #GTA #SysAdmin #DevOps #ZeroTrust

  2. Viele binden LLMs wie Web-Apps von 2002 an: User-Input rein, Modell baut SQL, ab in die DB.

    Front-End-Sanitization nützt nichts bei Indirect Prompt Injection, wenn externe Payloads (Mails, PDFs) unbereinigt im System-Prompt landen.

    Niemals stochastischen Output ungeprüft ausführen: Nutzt AST-Parser, Read-Only DB-Rollen & strikte Schemas.

    Wie kapselt ihr LLM-Tools ab?

    #AppSec #InfoSec #DevSecOps #Python #WebDev

  3. Many teams build LLM connectors like web apps in 2002: grab user input, let an AI generate a query, execute it raw against the database.

    Front-end sanitization won't save you from Indirect Prompt Injection when untrusted payloads (emails, PDFs, scraper) hit your system prompt.

    Never execute raw stochastic outputs. At minimum: AST parsing, read-only DB roles, strict output schemas.

    How are you sandboxing LLM tool-call in production?

    #AppSec #InfoSec #DevSecOps #Python #BuildInPublic #AIethics

Share on Mastodon

Enter the server where you have an account.