Outpoll | Prompt Security's Itamar Golan on Building a Generative AI Security Category

AIenterprise aiAI in Finance and Banking

The generative AI security landscape has rapidly evolved from theoretical concern to critical enterprise priority, a transformation vividly illustrated by Prompt Security's trajectory under co-founder and CEO Itamar Golan. VentureBeat's recent discussion with Golan revealed how his academic background in transformer architectures, long before they became the bedrock of modern large language models, provided unique foresight into the emerging vulnerabilities created by LLM-driven applications.His practical experience building one of the earliest GenAI-powered security features using GPT-2 and GPT-3 convinced him that these systems were creating an entirely new and democratized attack surface, where creativity alone could become an exploit vector. This foundational insight led to the August 2023 founding of Prompt Security, a $23 million funding raise across two rounds, the assembly of a 50-person team, and culminated in a strategic acquisition by SentinelOne for an estimated $250 million in August 2025—all within a remarkably compressed two-year timeframe.The urgency of this mission is underscored by stark data: VentureBeat analysis indicates shadow AI now costs enterprises $4. 63 million per breach, 16% above average, while IBM's 2025 data shows 97% of breached organizations lack basic AI access controls.The sprawl is accelerating, with VentureBeat estimating shadow AI apps could double by mid-2026 based on current growth rates, and Cyberhaven revealing that 73. 8% of ChatGPT workplace accounts are unauthorized.Golan's strategic decisions were pivotal in navigating this explosive growth. He deliberately chose to build an entire category—an AI security control layer—rather than compete on individual features like prompt injection defense.This involved embracing enterprise-grade complexity from the outset, developing for self-hosted and hybrid deployment models, and covering diverse surfaces from browsers to agentic workflows. Furthermore, he prioritized depth over customer volume, embedding with a select group of serious enterprises to shape a product that reflected real-world AI rollout challenges.A defining moment came from an incident with a large, regulated company that had deployed a customer-facing GenAI support agent. Despite a textbook security posture, a non-technical user, through carefully crafted natural language, managed to prompt-inject the agent into revealing sensitive customer data from other support tickets.This crystallized the paradigm shift: AI doesn't just introduce new risks; it democratizes them, compressing exploit discovery time and expanding the potential damage radius. This incident validated Prompt Security's core thesis and accelerated development of runtime protection for customer-facing apps and cross-tenant data leakage prevention.

#featured

#generative AI security

#prompt injection

#data leakage

#shadow AI

#SentinelOne acquisition

#enterprise adoption

#AI governance