In this talk Donato dives into concrete techniques we implemented in Spikee, our open-source LLM testing tool, to bypass production guardrails (for example Azure Prompt Shields and AWS Bedrock Guardrails). Drawing on two years of assessments, he highlights attacker strategies such as best_of_n and anti-spotlighting attacks and shows how Spikee encodes those approaches to probe prompt injection, data exfiltration, XSS and resource-exhaustion vectors across the full LLM application pipeline. Aimed at security testers and developers, the talk demonstrates how Spikee operationalizes these tests and surfaces actionable findings to harden real-world deployments.