Showing Posts From

October 2025

Evaluating LLM Input Comprehension and Guardrail Robustness through Noise-Based Attacks

We evaluate the ability of LLMs to understand text with random noise, and examine how prompts with varying levels of noise could bypass LLM guardrails.

Entra Connect Exploitation in 2025: An Overview

Entra Connect is the bridge between Microsoft's on-prem and cloud worlds, synchronising Active Directory and Entra ID identities. As such, it has traditionally served as a high-value target for threat actors. Conversely, continuous imrpovements by Microsoft have drastically changed the attack surface, leading to confusion about which exploitation techniques are still relevant and which aren't. This post will aim to clarify and document the Entra Connect exploitation landscape in 2025, and what defenders need to know to stay ahead.