AI in production changes who owns the risk
As AI systems begin operating in live environments, ownership of risk starts to shift.Rony Ohayon, CEO and Cofounder of DeepKeep, told ChannelE2E that responsibility is no longer confined to model builders or data teams.“As AI moves into production, responsibility becomes increasingly shared between security, data and compliance teams. The risk aspect falls heavily on the security team, because it shifts from affecting model performance to real operational exposure,” Ohayon said. “Once AI interacts with users, sensitive data, enterprise applications, and other AI agents, security teams should become the owners of the ongoing risk management.”
Filling the gap left by traditional DLP
Most enterprises already run mature data loss prevention tools, but those controls were built for structured data flows like email, file storage, and SaaS applications. They struggle to interpret free-form prompts, AI-generated responses, or multimodal interactions.Ohayon explained where DeepKeep fits into that picture. “Traditional DLPs remain valuable because they protect data at rest and in transit using deterministic rules. However, they were never designed to understand advanced LLM conversations, multimodal prompts, or semantic context,” he said.DeepKeep positions itself as an AI security control plane that sits between users, agents, applications, and AI models, inspecting interactions in real time. “DeepKeep fills this gap by sitting between users, agents, apps, and any GenAI model as an AI security control plane that evaluates prompts and responses at run time,” Ohayon said. “By governing the interaction layer, where sensitive data and AI reasoning intersect, DeepKeep ensures safe and compliant AI usage across enterprise workflows.”Rather than simply alerting after the fact, the platform is designed to enforce runtime actions that reduce risk as AI systems operate.Signals that push enterprises toward AI-specific controls
Adoption of AI security tooling is not driven by a single factor. DeepKeep is seeing a mix of regulatory pressure, internal assessments, and real-world incidents shaping demand.“We see several triggers driving AI security adoption, varying from regulation to organizations experiencing real incidents or reading about them,” Ohayon said. “One of the signals is when an organization starts moving from the experimental phase of using AI into implementing it into its organizational processes.”
“Some of the patterns we’re seeing include the need for a broad solution, the ability to enforce a cross-model unified policy, and flexible deployment options such as private clouds and on-prem environments.”
The PII guardrail reflects a broader shift in how companies think about AI security. Instead of adding protections after problems appear, privacy controls are built directly into everyday AI workflows. This allows security, IT, data, and governance teams to work from a shared layer of control rather than separate tools and processes. As AI use continues to expand, security tools need to understand context, apply policies consistently across models, and run continuously in the background. Capabilities like these make it easier for organizations to scale AI use while staying compliant and reducing risk over time.
