Get 15% off for Cybersecurity Awareness Month. Use code CYBERREADY15. Explore ideas

AI & Machine Learning Threat Modelling Workshop

From compromised training datasets to manipulated inputs, vulnerable models and unreliable outputs, understanding where threats can arise is essential to developing secure AI systems.

In this hands-on workshop, your team explores a realistic AI/ML system architecture, uses the Elevation of MLSec card game and the DIMO threat modelling framework, to investigate potential threats and learns a structured, collaborative approach to AI security risk identification.

Book a workshopExplore the workshop kit

  • Who should attend

    This workshop is designed for people involved in developing, deploying, managing or securing AI and machine learning systems.

    It's particularly relevant to:

    • AI and machine learning engineers
    • Software developers
    • Data scientists
    • Application security and DevSecOps professionals
    • Security architects
    • AI security consultants
    • Security Champions
    • Technical leads

  • What your team will takeaway

    By completing the workshop, participants gain:

    • Practical experience identifying AI and machine learning security threats.
    • Familiarity with the DIMO threat categories.
    • Experience analysing an AI/ML architecture from a security perspective.
    • Examples of documented security threat scenarios.
    • Greater confidence discussing AI security across technical teams.
    • A repeatable approach for starting AI threat modelling exercises.

  • Professional Skills Development

    The workshop develops practical competencies in AI and machine learning security, threat identification, risk analysis and collaborative security assessment.

    These learning activities are relevant to professional development in cybersecurity, particularly security risk assessment and the evaluation of emerging technology threats.

    We are developing additional curriculum mapping and learning documentation to support participants who wish to evidence relevant Continuing Professional Education (CPE) activities.

    Contact us to find out more

Workshop at a glance

  • Duration: Two hours
  • Delivery: In person or live online
  • Methodology: DIMO - Dataset, Input, Model and Output Risks
  • Learning format: Facilitator-led, interactive threat modelling game
  • Audience: AI/ML engineers, developers, security professionals and technical teams
  • Prior experience: No previous threat modelling experience required
  • Materials: Elevation of MLSec cards, a sample AI/ML architecture, training exercise and threat-recording resources
  • Facilitation: CyberSec Games or an authorised workshop facilitator

How it works?

  • Overview

    • Facilitator: CyberSec Games or Adam Shostack
    • Duration: 90 to 120 minutes
    • In-person or Zoom
    • Group size: Teams of 5 to 7 players
    • Session materials: For both virtual and in-person sessions, each player will receive an Elevation of MLSec deck, a training exercise and a copy of the system architecture diagram.

  • Before the session

    For virtual sessions, random card hands are generated using the online Croupier tool and emailed to players ahead of the session. Players pick out their cards from their deck before we start.

    Scorekeeping: Either someone from CyberSec Games or a nominated team member keeps score. If it’s a client team member, we'll provide a short pre-session briefing.

  • Session details

    The session begins with a short introduction to the game and the case management architecture scenario we’ll be using.

    Then we jump into live gameplay: players take turns applying their cards to the scenario, identifying threats, discussing risks, and capturing insights.

    We finish with a group discussion on what we observed and practical ways to introduce the approach into your own teams.

Interested in booking a session for your team?

If you're thinking about bringing this session to your team, drop us a line. We're happy to chat about what you need, talk through possible dates, and help figure out if it's the right fit.

Contact form

Frequently Asked Questions

What is Elevation of Machine Learning (ML) Security?

Elevation of Machine Learning Security (MLSEC) is a threat modelling card game designed to help teams identify and discuss security threats affecting AI and machine learning systems.

Created by Elias Brattli Sørensen, the game draws on security research from the Berryville Institute of Machine Learning (BIML) and OWASP.

It turns complex security risks into practical prompts for collaborative threat modelling.

What is AI Threat Modeling?

AI threat modelling is the process of systematically identifying potential security threats affecting AI and machine learning systems.

It involves examining components, data flows, dependencies and possible attack scenarios to understand how a system could be compromised or misused.

The aim is to identify threats early enough to inform security design and risk management decisions.

How is AI threat modelling different from traditional threat modelling?

Traditional application threat modelling commonly focuses on software components, data flows, trust boundaries and established security threat categories.

AI and machine learning systems introduce additional considerations involving training datasets, models, inference inputs, outputs and machine learning pipelines.

Elevation of MLSEC helps teams explore these AI-specific risks using the DIMO framework.

What is the DIMO framework?

DIMO stands for Dataset, Input, Model and Output Risks.

These four categories organise the security threats explored by Elevation of MLSEC, helping participants systematically consider different parts of an AI or machine learning system.

Do I need experience in AI security to participate?

No previous AI security or threat modelling experience is required.

The workshop introduces the DIMO categories and guides participants through the process of identifying threats.

Some familiarity with software development, AI or machine learning concepts will be beneficial.

Is this workshop suitable for generative AI and large language models?

The game includes threats relevant to AI and machine learning systems, including LLM-specific risks informed by OWASP research.

However, this is a general AI/ML threat modelling workshop rather than a comprehensive assessment of every threat associated with generative AI, RAG or autonomous AI agents.

Workshops involving particular AI architectures may require additional scenarios or specialist exercises.

Do participants need physical cards for online workshops?

Yes and we will ship the decks to the participants as part of the package. 

How is the workshop delivered online?

We've created a way to experience the session online without losing the tactile, interactive experience of the game.

In a traditional in-person workshop, participants draw cards from a shared deck. Our application Croupier brings this process online by managing randomised card hands for participants.

Each player receives their hand by email with the help of Croupier and picks out the matching cards from their deck ready for the session.

What’s the architecture scenario we’ll be threat modelling?

The architecture we have prepared for the Elevation of MLSec session contains a plausible deployment of several AI components including machine learning for image recognition, a pair of chat bits connected to a large language model and an MCP server to enable decisions to be executed.

How much does the session cost?

Pricing depends on the number of attendees and the delivery format (virtual or in-person). We’ll provide a clear quote once we know your team’s requirements.
Back to top