CloudShield AI Get in touch
Grand Prize · Craft HackAIthon 2024

CloudShield AI

AI-Driven Defense Against Cloud Vulnerabilities

Multiple LLM agents that autonomously find, exploit, and remediate cloud security vulnerabilities — in real time.

Piergiuseppe Mallozzi and the CloudShield AI team accepting the $10,000 grand prize from David Sacks at HackAIthon 2024.
Accepting the $10,000 grand prize from David Sacks (Craft Ventures) at HackAIthon 2024.

Why It Matters

Cloud security is a critical challenge in today's digital landscape, with misconfigurations and cyber attacks causing billions in losses annually. Security engineers must manage thousands of access policies for countless users, groups, roles, and data stores, making the system prone to errors. Frequent changes, such as adding temporary contractors, complicate tracking and verification. AWS alone offers thousands of actions, providing numerous opportunities for attackers to exploit. New strategies emerge regularly, further exacerbating these vulnerabilities.

  • Slow Detection

    With data stored in the cloud, it takes an average of 277 days to identify and contain a breach.

    IBM Report
  • Increased Attacks

    Attacks on open-source libraries, critical to cloud systems, have increased by 742% since ChatGPT's release.

  • Phishing Attacks

    Phishing emails, which can target cloud credentials, have increased by 1265% since ChatGPT launched.

    CNBC Report

Our Vision

We aim to leverage AI and LLM agents to provide an autonomous, intelligent defense system capable of dynamically analyzing current permissions, identifying and exploiting potential vulnerabilities in real-time, and offering detailed remediation strategies to enhance cloud security policies and prevent potential attacks.

Technical Framework Overview

CloudShield AI runs on a multi-agent architecture that integrates Tasks, AI Agents, and Tools. Tasks define the steps for identifying data access paths, classifying documents, and simulating attacks. Specialized AI agents execute these tasks both sequentially and asynchronously, autonomously coordinating and invoking tools that interact with cloud APIs and external packages (e.g. for encryption). This architecture ensures validated actions, minimizing inaccuracies and “hallucinations.”

Diagram of the CloudShield AI framework: tasks, AI agents, and tools interacting with cloud APIs.

AI Agents Workflow

Our AI agents are given broad task descriptions in natural language. Each agent autonomously creates a sequence of steps, calling the appropriate tools in the correct order. If an exception is raised from the tools, the agent learns, recovers, and attempts the task again.

Analyze Access Paths
Classify Content
Perform Attacks
Collect, Analyze, and Report

The Team

Piergiuseppe Mallozzi

Piergiuseppe Mallozzi

AI and Cyber-Security Researcher

Watch our Demo.

Get in touch

Want to collaborate or learn more about CloudShield AI? Reach out and let's talk.