Webinar Description
Key Takeaways
- Virtual webinar examining security challenges specific to AI agent architectures
- Covers hooks, harnesses and secure coding practices for autonomous AI systems
- Designed for security professionals, AI developers, DevSecOps teams and technical leaders
- Addresses threat modelling, vulnerability management and compliance for AI deployments
- Relevant to technology, finance, healthcare and enterprise IT sectors adopting AI at scale
Introduction
Coding Agents, Hooks & Harnesses: Securing the Next Generation of AI Agents is a virtual webinar bringing together security practitioners and AI specialists to examine the distinct challenges of protecting autonomous AI systems. As organisations increasingly deploy AI agents capable of executing code, accessing external services and making decisions with minimal human oversight, the attack surface expands in ways traditional application security frameworks were not designed to address. This event provides technical guidance for professionals responsible for securing these emerging architectures.
About This Event
The webinar offers expert-led sessions and interactive discussions focused on the security implications of AI agent development and deployment. Attendees will explore how hooks and harnesses function within agent architectures and why these components introduce specific vulnerabilities that require dedicated security controls. The format emphasises practical knowledge transfer rather than theoretical discussion, with content calibrated for professionals already working with AI systems or preparing to integrate them into production environments.
Securing Autonomous AI Architectures
AI agents differ fundamentally from conventional software applications. Where traditional programs follow predetermined logic paths, agents operate with degrees of autonomy that allow them to interpret instructions, select tools and execute multi-step workflows. This autonomy creates security considerations that extend beyond input validation and access control. Hooks—the integration points where agents connect to external systems—represent potential vectors for injection attacks, data exfiltration and privilege escalation. Harnesses, which provide the scaffolding for agent execution, must enforce boundaries that prevent agents from exceeding their intended scope.
The webinar examines how security teams can implement controls at each layer of the agent stack without undermining the flexibility that makes these systems valuable. This includes techniques for constraining agent behaviour, monitoring execution patterns and detecting anomalous activity that may indicate compromise or misuse.
Threat Modelling for AI-Driven Systems
Conventional threat modelling methodologies require adaptation when applied to AI agents. The non-deterministic nature of large language model outputs, combined with the potential for prompt injection and indirect attacks through retrieved content, introduces threat categories that security professionals may not have encountered previously. The event addresses how to construct threat models that account for these characteristics while remaining practical enough for integration into existing security programmes.
Particular attention is given to supply chain considerations, as AI agents frequently depend on external models, plugins and data sources that may themselves be compromised or manipulated. Understanding these dependencies is essential for accurate risk assessment.
Integrating Security into AI Development Lifecycles
For organisations building AI agents internally, the webinar explores how security practices can be embedded throughout the development process rather than applied retrospectively. This encompasses secure coding standards specific to agent development, testing methodologies that account for emergent behaviours, and governance frameworks that ensure compliance with evolving regulatory expectations around AI systems.
DevSecOps teams face particular challenges when working with AI components, as traditional static analysis and penetration testing techniques may not surface vulnerabilities that only manifest during agent execution. The event discusses complementary approaches that address these gaps.
Who Should Attend
The webinar is designed for security professionals, AI developers, DevSecOps practitioners, CISOs, CTOs and security architects working in organisations that are deploying or developing AI-driven systems. It holds particular relevance for those in technology, financial services, healthcare and enterprise IT, where AI adoption is accelerating and the consequences of security failures carry significant operational and regulatory implications. Attendees should have foundational knowledge of application security concepts, though deep expertise in AI is not a prerequisite.