Miracle Owolabi
Cyber Security Professional
Cybervergent
Cybersecurity professional specializing in red teaming AI systems, adversarial ML, and enterprise threat simulation. Focused on offensive security research and practical defense of modern machine learning pipelines.
Speaker sessions
Red Teaming the AI-Augmented Enterprise: Exploiting the Hidden Risks in Modern ML Pipelines
As enterprises rapidly integrate AI/ML into business operations, traditional pentesting fails to address emerging risks in model-driven systems. This talk presents a field-tested red team methodology tailored to modern ML pipelines, focusing on attack surfaces like LLM-based interfaces, embedding search, retrieval-augmented generation (RAG), model registries, and feature stores.Through real-world offensive engagements, I’ll walk attendees through how we discovered and exploited hidden vulnerabilities: prompt injections that pivoted into internal tools, data poisoning that silently corrupted fraud detection logic, and adversarial inputs that bypassed content moderation.The session dives deep into offensive TTPs adapted for AI targets, such as chaining insecure plugin calls, abusing misconfigured vector stores, and manipulating model training inputs. I’ll also cover responsible disclosure, testing safeguards, and aligning with MLOps and data governance teams.Attendees will leave with a red teaming playbook for AI systems, actionable checklists, and insight into how modern enterprises are building (and breaking) AI trust boundaries in production environments.
- 18:00
- Tue
- 02 Dec
Stage:
Briefings 2
Sessions Type:
Presentation