Talks

The AISVS project leads walk through scoping an assessment, choosing verification levels and applying the requirements to LLM apps, agents and MCP-connected tools.

Applying chaos engineering to AI systems: experiments that mimic realistic failures and adversarial scenarios such as tampered data, prompt injection and degraded models.

Turning AI security into testable, verifiable requirements, and how teams can apply AISVS from design through production readiness.

As agents gain memory, their context becomes an attack surface: altered history, biased knowledge and dormant instructions that fire later. Why agent memory must be protected like critical infrastructure.

AISVS tour: structure, three verification levels, and how it relates to ASVS, NIST AI RMF and ISO 42001.

How agent memory gets corrupted over time – rewritten history, biased knowledge, dormant instructions – and why it must be defended as critical infrastructure.

Hands-on lab with the AISVS project leads: scope an AI security assessment and apply AISVS to LLM apps, agents and MCP-connected tools.

The follow-up to Barcelona: breaking and defending agentic AI – MCP, A2A, memory poisoning, tool misuse and goal hijacking – plus Zero Trust-style isolation, sandboxing and runtime plan validation.