Mock-up of the TRACE interface: control centre, investigation and agents

Case study — ESSEC hackathon · AI-agent security

2026

Secure Horizons

01The project

Year
2026
Category
ESSEC hackathon · AI-agent security
Stack
  • Python
  • Next.js
  • Supabase
  • PostgreSQL
  • OpenAI API
  • RAG

36 hours on AI-agent security and human oversight, with KPMG mentors: a SOC copilot to investigate incidents involving compromised agents.

ESSEC Secure Horizons hackathon: how to keep control over AI agents that run tools, and investigate when one is compromised?

My role

Project lead: task allocation, scoping of the user journey, trade-offs over 36 hours and presentation to the KPMG mentors.

02Approach

  1. An investigation interface (TRACE) wired to a RAG assistant on Supabase and a Python defence pipeline.

  2. A gateway that traces agent conversations, a vault, decoys and an audit of installed packages; tool suspensions are applied after human approval.

03Results

  • End-to-end prototype: question → RAG → traces → incidents → containment

  • Python tests passing, front-end build green

Mock-up of the TRACE interface: control centre, investigation and agents
Fig. 01Mock-up of the TRACE interface: control centre, investigation and agents

Next project

BizroastSaaS · AI business analysis — 2026