From Weak Keys to AI Agents: How Machine Identity Became the New (Yet Old) Attack Surface

Abstract:

Machine identities (certificates, keys, tokens, and other cryptographic artifacts) are the backbone of secure digital interactions. They enable API calls, service-to-service communication, workload authentication, IoT onboarding, and encrypted data exchange.

Once rare, machine identities now vastly outnumber human accounts, and their growth is accelerating. The rise of AI agents is amplifying this trend, with 82% of global organizations expected to integrate agents within the next 1–3 years. Yet, cybersecurity practices have struggled to keep pace. Gartner predicts that by 2028, one in four enterprise breaches will stem from AI agent abuse.

This session traces the evolution of machine identity exploitation, from weak random number generators in early Unix systems to modern attacks on compromised service accounts. We’ll examine pivotal failures, explore how security techniques have advanced (including certificate-based authentication, hardware root of trust, ephemeral identities, SPIFFE IDs, MCP/A2A protocols), and discuss why securing machine identities today is both easier and more complex than ever before.