As enterprise AI adoption skyrockets, a dangerous new blind spot has emerged: the perception that AI components are secure by design. However, while your network and application infrastructure might be protected, the AI models and agents living within it are inherently insecure and quickly emerging as the newest weakest link in your cybersecurity posture. Join this hands-on workshop to challenge your assumptions and expose the hidden insecurities in the LLM and Agentic AI landscape. We’re moving beyond theory to demonstrate high-priority threats—including Prompt Injection, Data Poisoning, and Model Inversion—that traditional security controls often miss or simply are not equipped to address.
What to Expect:
- Live Attack Simulation: Engage in an interactive scenario where we simulate a corporate AI application compromised using natural language to "jailbreak" its guardrails, demonstrating how easily an attacker (or an unwitting insider) can trigger unauthorised data exposure, exfiltration and privilege escalation.
- Defence in Depth: Share your thoughts in real-time on appropriate defence measures to discover which platform-led approaches provide real-time protection through:
- Prompt Sanitisation: Neutralising high-risk instructions before they reach the AI application model.
- Contextual Anomaly Detection: Identifying (and blocking) subtle patterns behind sophisticated jailbreaks attempts.
- I/O Validation: Ensuring model responses adhere to strict policy, preventing the leakage of system prompts and internal data. Protect your “AI Achillies Heel”.
Learn how to build a robust defence strategy that protects your organisation from the unique risks of AI models and agents and empower transformative AI adoption responsibly and securely.
This session is suitable for Network Security Professionals, Security Architects, SOC Managers/Engineers, AI Security Architects, Security Engineers, DevSecOps.
Particpants will gain a first hand understanding of the key tenant of model-specific vulnerabilities and attack vectors within AI applications.
