AI projects rarely fail because of the model itself.
They fail because the supply chain around it is weak.

In this session, we show how to secure the entire supply chain of your AI solutions.
From data sources and training pipelines all the way to deployment and operations.

We highlight typical risks such as data poisoning, manipulated models (“PoisonGPT”), strong cloud dependencies, compliance requirements (e.g. GDPR), and IP theft.
Along a concrete value chain, we discuss:

Data Governance & Lifecycle Management
Collection, classification, storage, versioning, and archiving of data.

Secure Training Pipelines
Verified models, hash and signature checks, reproducible MLOps pipelines, isolated environments.

Secure Development for AI
Secure dev workspaces, SBOMs for ML stacks, test setups for agent-based and LLM-driven systems.

You will leave with concrete building blocks and architecture principles.
So you can design AI projects that are secure, auditable, and ready for production.
From first requirement to final delivery.

Session language: German

Target audience:
- Cloud and software architects
- Security and compliance leads
- Data & AI engineers / MLOps engineers

Technical level: 200 – architecture and process focused.

Format & duration (flexible):

25 minutes talk + 10–15 minutes Q&A

Speakers:
Rico Komenda, Senior Consultant, adesso SE
Marc Iridon, Senior Consultant, adesso SE

Requirements / setup:
Projector or large screen (HDMI input)
Audio output for short demo or video snippets (optional)

Special notes:
Product neutral, but with examples from the Microsoft ecosystem (e.g. Responsible AI, cloud security services).

Focus on pragmatic measures you can implement in existing cloud and enterprise environments.

In a nutshell:
An end-to-end view on securing AI supply chains.
With clear, practical patterns for data, training, development, and operations.