AI Devs 2
Hands-on training in building LLM applications and agents. Building them is what makes their failure modes obvious — which is most of the value for security work.
Updated
- Prompt design and structured output
- Retrieval-augmented generation pipelines
- Agents, tool calling and orchestration
- Evaluation and cost control
Why I took a builder's course, not a security one
AI Devs 2 is a Polish programme on connecting GPT-4 to application and automation logic — RAG pipelines, tool-calling agents, orchestration, cost and evaluation. It's not a security course, and that's the point: reviewing an AI feature for prompt injection or excess agent authority is a different exercise once you've built the retrieval pipeline and the tool-calling loop yourself and watched where they break on their own. Most of what I use from it now is the same threat modeling I run on any other engineering system, applied to a component that fails in less familiar ways.
Related
AI security
How to secure GenAI, LLM and agentic AI systems: current OWASP threat classes, layered controls, agent authority, retrieval isolation and monitoring.
AI security assessment
Security assessment of GenAI and LLM systems: prompt injection, tool and agent abuse, data leakage, model supply chain and monitoring gaps.
LLM security controls
A control checklist for enterprise GenAI: data handling, retrieval isolation, agent authority, cascading-failure containment, monitoring, cost limits and vendor terms.
Prompt injection defence
Why prompt injection cannot be filtered away, why it has no full fix yet, and how to design AI systems — architecturally, not just with better prompts — so a successful injection has no useful consequence.
Tokenization and filter bypass
Content filters match on words; models read tokens. Spacing, encoding and hidden characters change tokenization and can slip content past the filter.
Prompt injection
An attack where text processed by a language model is interpreted as instructions, overriding the developer's intent.