The cybersecurity landscape is currently undergoing a structural transformation that mirrors the chaotic, high-stakes maturation of the early fraud detection market. According to Jonathan Care, Lead Analyst and Practice Lead for AI at KuppingerCole Analysts, the current frenzy of AI security startups is a precursor to a massive, inevitable wave of industry consolidation.
In a wide-ranging interview, Care—who previously played a pivotal role at Gartner in defining the fraud management sector—outlines a future where the current "AI security" buzzword is stripped away, leaving behind a streamlined infrastructure of autonomous agent governance and identity-centric controls.
The Parallels of Innovation: From Fraud to Agentic AI
To understand where the AI security market is heading, one must look at the history of the fraud industry. Two decades ago, "fraud" was not a unified product category; it was a fragmented problem siloed between payment teams, risk departments, and IT. No single entity owned the budget, leading to a patchwork of point solutions—device fingerprinting, basic rules engines, and rudimentary behavioral analytics.
"Fraud was never a product category at the start," Care explains. "It was a problem that sat between the payments team, the risk team, and IT, and nobody owned the budget. Vendors sold point solutions. What turned it into a market emerged when buyers realized they needed to orchestrate those signals and decide in real time, at the point of transaction."
Care sees a direct correlation between that era and today’s Agentic AI security market. Currently, organizations are buying discovery tools, prompt filters, and AI posture management systems that function in isolation. He argues that the category will only mature when a centralized, orchestration-heavy approach—a "fraud decisioning engine" for AI—emerges. This system must make runtime decisions about what an agent is allowed to do, backed by deep context regarding the human user the agent represents.
The Lifecycle of a "Feature"
Care predicts a brutal culling of standalone AI security tools. "What becomes a feature?" he asks. His list includes model scanning, AI bills of materials (AI-BOM), and the bulk of AI security posture management (ASPM). These tools are destined to be absorbed into cloud security platforms, data security frameworks, or developer pipelines.
The market has already signaled this trend. The recent acquisitions of Astrix by Cisco, Apono by 1Password, and Cyata by Check Point represent a definitive move to fold machine identity and agent security into larger, more robust platforms. Standalone prompt-injection detection is, according to Care, "on borrowed time," destined to be subsumed by AI gateways and model providers’ native guardrails.
Market Dynamics: Q3 Funding and the "AI Label" Paradox
The economic landscape for security startups is increasingly polarized. Q3 funding data shows a 50% jump in capital deployment compared to the previous quarter, yet the total number of deals has plummeted. This creates a challenging environment for founders who lack the "AI" designation on their pitch decks.
Care notes that while capital is concentrating in fewer, larger rounds, the situation for non-AI founders is not hopeless—it is simply more rigorous. "Investors are pattern-matching on AI because that is where they believe the exits are," Care observes. "A founder without the label is fighting the pattern, so they need evidence instead."
The "Proof-Over-Vocabulary" Strategy
Care’s advice to founders is pragmatic: avoid the temptation to pivot to "AI" just to secure a meeting. Investors are increasingly sophisticated in their due diligence, and "AI-washing" can cost a startup its credibility. Instead, successful founders should focus on the unglamorous, foundational infrastructure that makes AI deployments possible.
"If you reduce losses, cut analyst hours, or close an audit finding, say so with numbers," he advises. The most significant opportunities lie in the "boring" sectors: identity management, logging, data handling, and third-party risk. By solving these, a company becomes an essential component of the AI stack, regardless of whether it labels itself as an "AI company."
The Israeli Ecosystem: Strength or Monoculture?
The role of Israeli startups in the global cybersecurity market has become a focal point of debate. With Q3 figures showing that Israeli firms secured roughly 33% to 46% of total global cyber venture investment—a staggering sum of $2.425 billion—the influence of the region is undeniable.
A Double-Edged Sword
Care views this concentration as both a pillar of strength and a potential risk. The strength lies in the unique pipeline of talent; many Israeli founders have deep, hands-on experience in cybersecurity, often having learned the craft from the "attacker’s side" during their military and professional tenures.
However, the risk is monoculture. When a vast majority of capital flows through a single ecosystem, it risks homogenizing the go-to-market playbooks and the types of problems being solved. "You see the same founding patterns, the same go-to-market playbooks, and often the same investors on the cap table," says Care. This can lead to a bias toward technically deep, offense-informed products while overlooking the patient, complex requirements of regulated industries in Europe and Asia. Furthermore, as critical infrastructure providers grow more cautious, the geographic location of engineering and support teams is becoming a factor in vendor continuity assessments.
The Five Commandments of Autonomous Production
When asked to strip away marketing fluff and define the minimum security requirements for an agent entering a production environment, Care provides a clear, actionable checklist. He insists that organizations must stop viewing agents as application features and start viewing them as identities.
Before an agent touches production, it must pass these five gates:
- Identity Ownership: Every agent must have a unique identity, a defined owner, a specific purpose, and a managed lifecycle. Shared service accounts are strictly forbidden.
- Granular Least Privilege: Access should be granted on a task-basis with short-lived, scoped credentials that expire immediately after the operation.
- Human-in-the-Loop: Any high-impact or irreversible action—such as deleting data, moving financial assets, or changing permissions—must require explicit human approval.
- Actionable Observability: Full logging is required. The system must record not only the agent’s request but the rationale behind its decision and the final outcome, in a format intelligible to human incident responders.
- Verified Kill-Switch: Organizations must have a functional, tested "off switch" that can immediately neutralize an agent gone rogue.
"None of this is exotic," Care notes. "It is identity governance and change control applied to software that makes its own decisions. The organizations that struggle are the ones that skipped those disciplines for humans and are now discovering they need them for machines."
The 2028 Forecast: Bets and Folds
As the market approaches 2028, Care offers a glimpse into what he believes will thrive and what will fade into the background.
Betting On:
- Runtime Authorization for Agents: Systems that provide real-time, context-aware policy decisions at the exact moment an agent performs an action.
- AI-TDR (Identity Threat Detection): A focus on non-human identities as primary security citizens, given that the population of machine identities is exploding relative to human users.
- Outcome-Based AI-SOCs: Tools that measurably reduce analyst workloads and provide auditable outcomes rather than "black box" demonstrations.
Betting Against (As Standalone Categories):
- Prompt-Injection Filtering: destined to become a standard feature within model platforms and gateways.
- AI Security Posture Management (ASPM): Expected to be absorbed entirely by existing cloud and data security platforms.
- "AI Firewalls": Inspection-only products that claim to make any model safe. Care argues that inspection is a secondary measure; safety must be engineered into the design of the system, not bolted on at the edge.
Ultimately, Care’s outlook is a reminder that the "AI revolution" in security is subject to the same laws of physics that governed previous technological shifts. The noise of venture capital will eventually quiet down, and the industry will be left with those who built the robust, boring, and necessary foundations of trust.
