Team Research blog
The IncludeSec team explores the practical considerations of AI-assisted source code analysis. Observing results produced with frontier vs locally-hosted models, various harness orchestration designs, and a list of different language frameworks provided us with meaningful insights into how the modern pentester should be using AI to produce higher quality results (but not necessarily in less time!)
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 attended the BSides 2026 CTF competition and observed how effective frontier models have become at solving a variety of CTF challenges. Our team gives a first-hand account of how the winning strategy has recently shifted to prioritizing efficient solve times and orchestration pipelines that can combine the speed of lighter-weight models on easier challenges with a heavy-hitting model’s reasoning capabilities. We also highlight why the success of LLMs in CTFs does not translate as effectively in the arena of professional security assessments.