
An Exclusive Conversation with Itamar Golan, CEO of Prompt Security
Recently, VentureBeat had the opportunity to virtually sit down with Itamar Golan, the co-founder and CEO of Prompt Security, to delve into the challenges posed by GenAI security for organizations of all sizes.
During our discussion, we explored topics such as shadow AI sprawl, the strategic decisions that led Golan to focus on building a leading platform rather than just competing on features, and a real-world incident that highlighted the crucial need for protecting AI applications. This conversation shed light on Prompt Security’s mission to empower enterprises to adopt AI securely, ultimately leading to its estimated $250 million acquisition by SentinelOne in August 2025.
Golan’s journey to founding Prompt Security began with his academic research on transformer architectures long before they became integral to today’s large language models. His experience in developing one of the earliest GenAI-powered security features using GPT-2 and GPT-3 convinced him of the new attack surface created by LLM-driven applications. Prompt Security was founded in August 2023, raised $23 million through two rounds of funding, built a team of 50 individuals, and achieved a successful exit in under two years.
The timing of our conversation couldn’t have been more opportune. Research by VentureBeat indicates that shadow AI now costs enterprises $4.63 million per breach, 16% above the average. Shockingly, 97% of breached organizations lack basic AI access controls, according to IBM’s 2025 data. Furthermore, it is estimated that shadow AI applications could double by mid-2026 based on current 5% monthly growth rates. Data from Cyberhaven reveals that 73.8% of ChatGPT workplace accounts are unauthorized, and enterprise AI usage has surged 61 times in just 24 months. As Golan previously shared with VentureBeat, “We see 50 new AI apps a day, and we’ve already cataloged over 12,000. Around 40% of these default to training on any data you feed them, meaning your intellectual property can become part of their models.”
The following has been edited for clarity and length.
VentureBeat: What made you recognize that GenAI security needed a dedicated company when most enterprises were still figuring out how to deploy their first LLMs? Was there a specific moment, customer conversation, or attack pattern you saw that convinced you this was a fundable, venture-scale opportunity?
Itamar Golan: From an early age, my passion for mathematics, data, and artificial intelligence led me down an academic path that culminated in studying transformer architectures. This interest in AI also shaped my career as a data scientist, where I increasingly encountered intersections between AI and cybersecurity.
Everything changed with the release of the first OpenAI API. During that time, I collaborated with Lior Drihem, who later became my co-founder and Prompt Security’s CTO. Together, we developed one of the earliest security features powered by generative AI, leveraging GPT-2 and GPT-3 to generate actionable remediation steps for security alerts. This significantly reduced the time required for security teams to address and resolve issues.
This experience made it evident that applications powered by GPT-like models were introducing a new and vulnerable attack surface. Recognizing this emerging risk, we established Prompt Security in August 2023 to address these challenges. Our objective was to empower organizations to embrace AI innovation without compromising security and governance.
Prompt Security gained recognition for its prompt injection defense, but you were addressing a broader spectrum of GenAI security challenges. Can you walk us through the full scope of the platform’s capabilities, including data leakage, model governance, compliance, red teaming, and any surprising features that resonated with customers?
Right from the start, Prompt Security was designed to cater to a wide array of use cases. Our focus extended beyond mere employee monitoring or prompt-injection protection for internal AI applications. To instill confidence in security teams regarding the safe adoption of AI, we needed to safeguard every touchpoint across the organization in real-time.
For many customers, a pivotal moment came when they discovered the multitude of AI tools their employees were already utilizing. Early on, companies often unearthed not just ChatGPT but numerous unmanaged AI services in active use, completely outside the purview of IT. This emphasized the critical nature of shadow AI discovery within our solution.
Equally essential was the automatic sanitization of sensitive data in real-time. Rather than outright blocking AI tools, we enabled employees to use them securely by removing sensitive information from prompts before they reached external models. This approach struck a balance, providing robust security without impeding productivity. Employees could seamlessly leverage AI while security teams were assured that no sensitive data was being leaked.
What surprised many customers was how facilitating safe usage, as opposed to restricting it, accelerated adoption and fostered trust. Once AI was perceived as a secure and managed channel rather than a forbidden territory, responsible usage surged.
You successfully positioned Prompt Security as a market leader. What were the strategic decisions that propelled your growth? Did focusing on a specific vertical play a role?
Looking back, the significant acceleration we experienced stemmed from deliberate decisions made early on. While these choices entailed discomfort, expense, and short-term setbacks, they ultimately generated substantial leverage over time.
Firstly, I opted to build a category rather than a mere feature. Right from the outset, I resisted positioning Prompt Security as merely offering protection against prompt injection or data leakage, recognizing the limitations of such a narrow focus.
Instead, I framed Prompt as the enterprise’s AI security control layer, governing interactions between humans, agents, and applications with LLMs. This foundational decision enabled us to establish a budget, assume a strategic role at the CISO table, implement platform-level pricing, and ensure long-term relevance instead of engaging in a feature race. Rather than striving to outdo competitors in features, I was pioneering a new category.
Secondly, I embraced enterprise complexity early on. While most startups shy away from complexity until compelled, I proactively designed for enterprise deployment models, including self-hosted and hybrid solutions. We addressed real enterprise interfaces such as browsers, IDEs, internal tools, MCPs, and agentic workflows, accepting longer cycles and more intricate engineering in exchange for credibility. This approach, though challenging, bestowed upon us a competitive edge that competitors could not easily replicate: enterprise readiness ahead of market demand.
Lastly, I prioritized depth over logos. Instead of pursuing volume or superficial metrics, I delved deep with a select group of serious customers, embedding ourselves in their internal AI deployment processes, risk assessment methodologies, policy frameworks, and long-term AI adoption strategies. These customers did not merely purchase the product; they actively shaped it. This collaborative approach resulted in a product that mirrored enterprise realities, generated compelling proof points that resonated with boardrooms rather than just security teams, and established a level of defensibility rooted in customer entrenchment rather than marketing strategies.
You were at the forefront of educating the market about threats that most CISOs had not yet considered. How did your positioning and messaging evolve from the inception of Prompt Security to its acquisition?
In the nascent stages, our focus was on educating a market grappling with the notion of AI adoption extending beyond a few employees utilizing ChatGPT for productivity. Our positioning primarily revolved around creating awareness among CISOs regarding the extensive reach of AI applications throughout their organizations and the consequent risks that had not been adequately acknowledged.
I wasn’t trying to win a feature race; I was building a new category.
As the market matured, our messaging transitioned from signaling the occurrence of AI sprawl to outlining strategies for staying ahead of the curve. CISOs now fully comprehend the magnitude of AI sprawl and acknowledge that basic controls or URL filtering are insufficient. Rather than debating the problem, they seek solutions that facilitate secure AI usage without the operational burden of tracking every new AI tool, site, copilot, or agent discovered by employees.
By the time of the acquisition, our positioning revolved around being the harbinger of safety: a solution that delivers visibility, protection, and governance at the pace of AI innovation.
Our research indicates that enterprises encounter challenges in obtaining senior management approvals for deploying GenAI security tools. How are security departments persuading C-level executives to forge ahead?
The most successful CISOs are framing GenAI security as a natural extension of existing data protection mandates rather than an experimental budget item. They portray it as safeguarding the same assets—corporate data, intellectual property, and user trust—in a burgeoning, rapidly expanding channel.
What’s the most significant GenAI security incident or near-miss you faced while developing Prompt Security that underscored the critical need for these protections? How did that incident influence your product roadmap or go-to-market strategy?
The pivotal moment that crystallized everything for me involved a prominent, highly regulated company that launched a customer-facing GenAI support agent. This initiative was not a haphazard experiment; the company had all the recommended security measures in place, including WAF, CSPM, shift-left practices, regular red teaming, and a secure SDLC process.
However, what they failed to fully anticipate was that the AI agent itself had become a vulnerable attack surface. Within a few weeks of its launch, a non-technical user discovered that by crafting specific conversational flows (not code or exploits, but natural language), they could prompt the agent to divulge information from other customers’ support tickets and internal case summaries. This breach was not orchestrated by a nation-state actor or a skilled hacker; it was essentially a curious individual with time and creativity. Yet, through a simple conversational interface, they accessed some of the most sensitive customer data held by the company.
This incident was both intriguing and alarming, revealing how creativity alone could serve as an exploit vector.
It was at that moment that I truly grasped the transformative impact of GenAI on the threat landscape. AI not only introduces new risks but democratizes them, rendering systems vulnerable to individuals lacking traditional hacking skills. It accelerates the discovery of exploits, compresses the time taken for breaches to occur, and significantly amplifies the damage radius once a breach transpires. This incident validated our original approach, propelling us to intensify our efforts in safeguarding AI applications beyond internal usage. Subsequently, we expedited work on:
• Providing runtime protection for customer-facing AI applications
• Detecting prompt injection and context manipulation
• Preventing cross-tenant data leakage at the model interaction layer
Furthermore, this incident reshaped our go-to-market strategy. Instead of solely emphasizing internal AI governance, we began illustrating to security leaders how GenAI transforms their customer-facing interfaces into high-risk, high-exposure assets virtually overnight.
What is your current role and focus following the acquisition by SentinelOne? How has operating within a larger platform company influenced your ability to innovate compared to running an independent startup? What aspects have become easier, and which have become more challenging?
Presently, the focus lies in extending AI security throughout the entire platform, integrating runtime GenAI protection, visibility, and policy enforcement within the same ecosystem that already secures endpoints, identities, and cloud workloads. While the mission remains unchanged, the scope has broadened.
Ultimately, we are working towards a future where AI itself becomes an intrinsic part of the defense framework—not merely something to secure, but something that fortifies your organization.
The Larger Landscape
Mergers and acquisitions are accelerating for GenAI startups that have demonstrated the ability to scale to enterprise-level security without compromising accuracy or speed. Palo Alto Networks recently acquired Protect AI for $700 million. Tenable’s acquisition of Apex Security amounted to $100 million, while Cisco reportedly purchased Robust Intelligence for $500 million. As Golan aptly pointed out, the companies that survive the forthcoming wave of AI-enabled attacks will be those that embed security into their AI adoption strategy from the outset.
Post-acquisition, Prompt Security’s capabilities will extend across SentinelOne’s Singularity Platform, encompassing MCP gateway security between AI applications and over 13,000 known MCP servers. Additionally, Prompt Security will provide model-agnostic coverage across major LLM providers, including OpenAI, Anthropic, and Google, as well as self-hosted or on-prem models as part of its integration into the Singularity Platform.
