Secure Use of AI in Software Development

The use of agentic AI is currently generating an enormous amount of new source code and self-developed applications in companies. Language models deliver more code in less time than a team can process through review and testing. At the same time, AI agents execute commands autonomously, install packages, and have access to sensitive data.

Alongside classic development teams, business departments now also program their own applications, often without following the established Software Development Lifecycle (SDLC). The productivity gains and the opportunity to optimize your own business processes through agentic software development, however, also bring risks. Leaked trade secrets, "hallucinated" malicious dependencies, and deleted production databases are possible consequences when the guardrails of the SDLC are not followed.

Within the workshop, we show you through live hacking demos how attackers can exploit AI models and AI agents, for example through prompt injection.

Together we walk through the typical Software Development Lifecycle and identify in each phase where agentic AI adds value, what risks arise, and what measures can secure its use.

For each phase, we work with you to develop actionable measures to effectively secure the use of agentic AI in your company. We support you in prioritizing and can also accompany you in the implementation after the training.

Training Goals

• See Through Attack Chains: You know the essential threats of AI-based attacks and risks in agentic software development.
• Develop Protective Measures: You can develop targeted protective measures in all phases of software development.
• Prioritize Measures: From the possible measures you derive a concrete, actionable list of quick wins and long-term hardening measures.
• Secure Use of Agentic AI: You can introduce agentic tools in a value-adding and secure way.

Course Content

• Lifecycle Security: You analyze the threat landscape across the phases of the Secure Software Development Lifecycle.
• Lethal Trifecta: For your AI use cases you identify the critical combination of data access, unsecured input, and exfiltration paths as the primary attack surface.
• Meta's Agents Rule of Two: You apply the principle of risk limitation by securing agents through restricting them to a maximum of two critical attributes.
• Agent Governance: You develop techniques for isolation (sandboxing) and centralized rights management.
• AI Attacks in Live Demos: You experience the use of manipulated prompts and the compromise of agents in practice.

Target Group

Requirements

Location

The trainings of the aramido Academy take place in the modern premises of aramido GmbH in Karlsruhe. You can reach us both by public transport and by car:

aramido GmbH
Durlacher Allee 75 – 77
76131 Karlsruhe

We are also happy to offer this training as in-house training or online.