Add to calendar Webinar

Securing AI agents: where threats hide in agentic systems

In this session, we break down how real-world agent attacks actually work, using actual payloads and documented breaches. Then we walk through the layered defences that hold up in production.

Share

Agenda

  1. Why AI agents are a new kind of security risk
  2. How attackers actually break AI agents
  3. What real-world breaches reveal
  4. How to defend agents: security pillars and layered guardrails
  5. Key takeaways and practical steps you can take starting now

Key topics

  • how granting an AI agent access to your tools, data and secrets changes the security model entirely;
  • the simple pattern behind almost every agent breach and why traditional security fails to detect it;
  • the surprisingly easy ways attackers manipulate agents, shown through real examples;
  • the real-world incidents that should give anyone deploying agents pause;
  • what actually works in defence and the first steps you can take right now.

Speaker

Piotr Krawiranda

Piotr Krawiranda

AI Architect

An AI Architect with 10+ years in engineering and 6+ focused on applied AI, he has designed and deployed systems across computer vision, deep learning, MLOps, and cloud-native AI architectures. He bridges research and production, from autonomous ground vehicles to enterprise applications.

Amazon Bedrock

Amazon Bedrock

The platform for secure and private agentic AI services

Enter the world of generative AI with Amazon Bedrock — a fully managed service offering top of the line foundation models, such as Anthropic's Claude and OpenAI's GPT.

Thanks to its serverless design, Amazon Bedrock eliminates the complexity of running AI at scale, and lets you seamlessly integrate agentic AI into your applications using familiar AWS services — whether it's your first steps with AI, or you are already revolutionizing the market.

Agentic AI services

AWS is trusted by…