Decoding Agentic: Securing the Data Layer AI Just Set on Fire

From Perimeter to Intelligence

Forcepoint’s Ronan Murphy tells Black Hat’s Mike Shema that security evolved from infrastructure to applications to identity—and now to protecting organizational “alpha”: proprietary data, IP, and regulated content feeding AI systems.

Agentic Expansion

Boards worry LLMs, copilots, and autonomous agents accessing SharePoint, OneDrive, and cloud repos may leak crown jewels into models or third parties. Murphy notes proliferation of non-human identities with broad repository access outpaces governance maturity.

Data Is Not CVEs

Unlike standardized vulnerability management, data value is contextual—a hospital’s sensitive records differ from manufacturing trade secrets. Organizations struggle to discover, classify, contextualize, and risk-score data before granting agents access.

Sanctioned vs Shadow AI

Even approved AI tools need least-privilege data access and prompt guardrails. Unsanctioned “shadow AI” is worse. Murphy describes rising executive concern—citing public commentary on Fortune 500 IP fears—and urges runbooks that let companies adopt AI while proving alpha protection to regulators.