We break your LLM features so your customers don't. 🛡️
Adversarial Testing | Red Teaming | AI Safety Audits |
EU ACT compliancy
Ex-Meta & Scale AI Specialists.
When I built the internal Red Team at Microsoft Azure Data ~12 years ago, there was almost no guidance on how to do it (hiring, measuring, planning,...)
So I wrote the book I wish existed.
6 years since release today. 🙌
Time flies...
https://t.co/CAbdRTEeIE
Politeness is not a security feature.🛡️
#Gemini is not yet EU ACT compliant
Target: [Gemini] ♊️
Violation: [Self-Harm] 🗡️
EU AI Act Status: Critical failure of Article 15 robustness standards.
#AIsafety#adversarialAttacks
Politeness is not a security feature.🛡️
The industry spent billions making LLMs helpful and harmless, but neglected their structural security. As 2026's EU Act kicks in, the "Safety Illusion" is finally breaking.
Read more 👇
https://t.co/949I1JLC9p
@wunderwuzzi23 Very interesting insight! So the Summarizer acts as an accidental Trojan? It "launders" a dormant injection into a fresh command!? Thats crazy; I am curious how they plan to fix this... #EUact