Team Research blog
What should security experts and their clients be considering when evaluating how AI is used in their security assessments? Why isn’t AI making all security testing cheaper? In this post, we discuss the tradeoffs of security testing that leverages AI.
Our team examines abliterated models and what scenarios in which security professionals should consider using them. This post briefly explains abliteration as a concept, discusses how it might make exploit development easier, and also discusses the risks of the current abliteration model landscape.
In this post we look under the hood of BrightData’s SDK and how it turns ordinary consumer TVs into exit nodes of an enormous commercial, residential proxy network leveraged by the AI industry to scrape web data and train language learning models.
The Include Security team takes a foray into the world of audio production equipment in our latest blog post. We look under the hood of a professional-grade audio mixer to explore its security profile, consider how its functionality could be leveraged by an attacker in a real world setting, and develop a proof-of-concept exploit to demonstrate quick n’ easy privilege escalation.
In this post we consider how to think about the attack surface of applications leveraging LLMs and how that impacts the scoping process when assessing those applications. We discuss why scoping matters, important points to consider when mapping out the LLM-associated attack surface, and conclude with architectural tips for developers implementing LLMs within their applications.