The Attacker’s Lens: The Hidden Path to Large-Scale LLM Exploits - Amir Shahmiri

The Attacker’s Lens: The Hidden Path to Large-Scale LLM Exploits As AI-powered agents rapidly expand across applications, they introduce a fundamentally new and largely unprotected attack surface. This talk explores how prompt injection evolves from a targeted technique into a scalable, internet-wide threat, where attackers embed malicious instructions across webpages, open-source content, ads, and shared knowledge sources to influence agent behavior at scale. By exploiting the growing capabilities of LLM agents, such as web browsing, tool usage, and autonomous decision-making, adversaries can trigger unintended actions including data exfiltration, privilege escalation, and system manipulation, often without direct interaction. We will examine why existing mitigations fall short, how these attacks mirror traditional OWASP categories in a new AI context, and what this means for the future of secure software design. Attendees will leave with a practical understanding of emerging AI risks and actionable strategies to design, test, and secure LLM-powered systems against this new class of threats. SPEAKER BIO: Amir Shahmiri, Senior Sales Engineer. Amir is an application security expert, with years of experience within the industry. He has worked predominantly within the application security space, specialising in helping organisation build safer and more efficient application security programmes. He is currently a Senior Sales Engineer with Mend.io, helping organisation build a mature, proactive AppSec programme that effectively manages application risk. This talk was presented at the OWASP London Chapter Meeting on March 31, 2026, an event dedicated to celebrating OWASP's 25th Anniversary. This event was hosted by ‪@tessl-ai‬ and sponsored by AVAQOR and ‪@Mend_io‬