The Terminology Problem Causing Security Teams Real Risks
Jailbreaks target the model's safety training; prompt injection hijacks application trust boundaries. Conflating them leads to defenses that miss your actual threat surface.
Fogel.dev's Blog
AI security research, open standards, and systems built from 0→1 by Ariel Fogel.
Latest posts
What Makes Indirect Prompt Injections Attacks Succeed
Anatomy of an Indirect Prompt Injection
Model Files Can Be Executables: A GGUF Case Study in Prompt-Template Backdoors
We reveal a verified supply-chain risk in GGUF: model files can execute hidden template logic. See a runnable demo and a 60-second manual check to review your template.
Design Priniciples for Developers
Design Priniciples for Developers
Dealing with Direct Upload Attachments in Tests
The process of directly uploading an attachment means that you are attaching a blob using the blob's `signed_id`. Read on to see how to access a blob's signed_id before the blob is attached to your ActiveRecord model instance in an RSpec suite.
How do Rails Form Helpers Know Where to Go?
Following `polymorphic_path` down the rabbit hole.
Ruby on Rails Podcast with Brittany Martin
Level Up in Learning with Ariel Fogel (that's me!)
What makes for a good pull request?
Wisdom from the distributed beyond
RESTful routes are still quite powerful
Are RESTful routes still a useful abstraction when we have other options?
HTML5 Canvas on Rails? Part 4
HTML5 Canvas on Rails? Part 4
HTML5 Canvas on Rails? Part 3
HTML5 Canvas on Rails? Part 3
HTML5 Canvas on Rails? Part 2
HTML5 Canvas on Rails? Part 2
HTML5 Canvas on Rails? Part 1
HTML5 Canvas on Rails? Part 1
Learning with Visual Representations (Part I)
This is a post on how to evaluate different representational competencies. Specifically, this post reviews Rau, M. (2016). Conditions for the Effectiveness of Multiple Visual Representations in Enhancing STEM Learning.