The AI security market is evolving, with some categories set to become integral features within larger platforms. Jonathan Care, Lead Analyst and Practice Lead AI at KuppingerCole Analysts, draws parallels between the developing AI security field and the early fraud market he helped define at Gartner.
Care points out that fraud initially wasn’t a distinct product category but a problem shared across payments, risk, and IT teams. Vendors offered point solutions, but the market solidified when buyers realized the need for real-time orchestration of signals at the transaction point.
Agentic AI Security: The Next Fraud Decisioning Engine?
Care sees a similar pattern in agentic AI security. Currently, the market offers discovery tools, prompt filters, guardrails, and AI posture management, each addressing a piece of the puzzle. He predicts that the category to emerge as a standalone product will be the one enabling runtime decisions about an agent’s actions, considering the context of who it represents — much like the fraud decisioning engine.
Model scanning, AI bills of materials, and most of AI security posture management will likely be integrated into cloud security, data security, or developer pipeline platforms. Standalone prompt-injection detection is also on borrowed time, as it will be absorbed into gateways and model providers’ guardrails.
This consolidation is already underway, with recent acquisitions like Cisco’s purchase of Astrix, 1Password’s acquisition of Apono, and Check Point’s move on Cyata, all aimed at incorporating agent and machine identity security into larger platforms.
Funding Trends and the AI Label
While Q3 funding increased by 50% compared to the previous quarter, the number of deals decreased across all stages. This shift favors larger rounds, making it harder for Series A founders without an AI label on their pitch deck. Care advises against artificially adding “AI” to attract investors, as it can damage credibility during diligence.
Instead, founders should focus on demonstrating tangible outcomes, such as reducing losses, cutting analyst hours, or closing audit findings, with measurable results. There’s also an opportunity for founders addressing the less glamorous aspects of security, such as identity, data handling, logging, and third-party risk, which are essential for AI deployments.
Israeli startups had a strong showing in Q3, raising at least $2.425 billion, nearly half of global cyber venture investment. While this highlights Israel’s strength in producing technically deep, offense-informed products, it also raises concerns about monoculture and the potential for overlooked problems that require patient work with regulated buyers in Europe or Asia.
For organizations deploying AI agents, Care emphasizes the importance of treating agents as a new kind of identity. He outlines five essential controls: unique agent identities, least privilege with short-lived credentials, human approval for significant actions, detailed logging, and a tested kill switch.
Looking ahead to 2028, Care identifies three categories he’d bet on: runtime authorization for agents, AI-assisted identity threat detection, and AI-assisted security operations that reduce analyst workload with auditable outcomes. Conversely, he predicts that prompt-injection filtering, standalone AI security posture management, and “AI firewalls” relying solely on inspection will become features within larger platforms.
Identity and Security Controls for AI Agents
First, each agent must have a unique identity, with a designated owner, purpose, and lifecycle. This prevents the use of shared accounts or human credentials. Second, the principle of least privilege should be applied, granting agents short-lived, task-specific credentials. Third, human approval is necessary for significant actions, such as data deletion or financial transactions.
Fourth, detailed logging is essential, capturing the agent’s instructions, decisions, and actions in a readable format for incident response teams. Lastly, a tested kill switch is key to ensure the agent can be promptly deactivated if needed.
Practical Resilience and Vendor Concentration
Buyers in critical infrastructure and government are increasingly scrutinizing the location of vendors’ engineering and support teams. This focus is not a reason to exclude Israeli vendors but rather a prompt to ask continuity questions typical of any concentrated supplier.
Future Bets and Predictions
The evolution of the AI security market mirrors the early fraud market, with consolidation and specialization shaping its future. Care’s insights offer a roadmap for founders, investors, and security professionals, emphasizing tangible outcomes, identity governance, and strategic category positioning.
Leave a Reply