Agentic AI Is Forcing a Rethink of Security Strategy
Security vendor acquisitions of AI infrastructure companies are aggressive moves by established cybersecurity players to build integrated defense stacks that can monitor, interpret, and control autonomous AI agents and non-human identities operating at machine speed across enterprise systems, because traditional tools built around human-centric events, static rules, and text filters cannot reliably manage agentic AI threats as these agents become production-critical in large organizations.
The headline story is not that security firms are buying promising AI startups; it is that they are admitting their old architectures are no longer enough. Cisco, A10 Networks, and others see that agentic AI workloads introduce a new class of risk tied to autonomous decisions, tool calls, and multi-modal execution traces rather than simple login events or traffic spikes. Instead of layering yet another point solution on top of brittle legacy stacks, they are racing to own the infrastructure layer that understands AI behavior itself. This consolidation trend is about control: whoever owns the telemetry and context for AI agents will define what “secure AI” means for the next decade.
Cisco’s WideField Deal: Building an Agentic SOC, Not Another Dashboard
Cisco announced its intent to acquire WideField Security, a company focused on security technology for the agentic AI era. WideField’s technology will be integrated into Splunk to enhance Agentic SOC capabilities, helping normalize and correlate identity, session, and activity telemetry from multiple sources. This is a clear statement that “AI infrastructure security” will sit inside the core observability and incident response stack, not as a bolt-on scanner.
Cisco explicitly ties the deal to the rapid deployment of AI agents, autonomous workloads, and non-human identities, which it says have introduced a new class of security risk. These systems can operate at machine speed, creating challenges around unauthorized access and around approved users or agents taking unsafe actions in the wrong context. By using identity telemetry and intelligence from a wide range of sources, Splunk’s Agentic SOC will be able to assemble session-level signals for deeper analysis by security analysts and distinguish legitimate sessions from potentially malicious ones. WideField also reinforces Cisco’s broader goal of delivering an integrated trust layer for the agentic AI era, spanning identity, runtime behavior, visibility, and enforcement. According to Kamal Hathi, WideField’s integration will help Splunk assemble context across human, non-human, and AI-agent activity, including signals from Cisco Identity Intelligence.
A10–TrojAI: Turning AI Threat Detection into Core Infrastructure
A10 Networks’ acquisition of TrojAI is framed around fusing AI-specific threat mitigation into the company’s existing infrastructure products, rather than selling yet another isolated AI scanner. The newly expanded security suite introduces native integration with the Model Context Protocol (MCP) to standardize visibility and access logs across interactive tool ecosystems, developer assistants like Claude Code, and local coding frameworks. This is infrastructure-level thinking: security must see what AI agents are doing in tools, memory, and databases, not only what they output as text.
A10 highlights its ability to map execution traces of multi-modal agents, supervising permission handshakes, system memory lookups, database extractions, and external tool execution instead of relying on standard text-filtering components. Adversarial vulnerabilities uncovered during automated build-time red-teaming cycles are configured to automatically feed discovery metrics back into A10’s proprietary guardrail models in near real time, creating a loop that hardens production-scale defenses without heavy client-side instrumentation. The unified product landscape fuses dedicated enterprise AI threat mitigation across A10’s Application Delivery Controller, distributed denial-of-service protection, web application firewall, and API security matrices to protect large-scale public sector and Fortune 50 corporate installations. A10 states that the cash transaction will not exert a material impact on its financial results for the 2026 fiscal year, positioning the integration to capture long-term enterprise demand for secure, data-sovereign AI infrastructure rollouts over the next 2 to 5 years.
From Point Products to Integrated AI Defense Stacks
Taken together, these security vendor acquisitions mark a decisive shift toward security consolidation trends in AI. Cisco’s WideField deal builds on its recent additions of Astrix Security and Galileo, reinforcing a strategy of an integrated trust layer for agentic AI spanning identity, runtime behavior, visibility, and enforcement. A10’s integration of TrojAI fuses AI threat mitigation with core ADC, DDoS, WAF, and API security offerings. Both moves show that the market will reward vendors that offer full-stack AI infrastructure security: telemetry, reasoning, enforcement, and guardrails living inside existing network and application layers.
Traditional approaches assumed humans at the center and focused on access control, signatures, and static rules. In an agentic AI era, the center is shifting to autonomous agents that call tools, read memory, and write code on behalf of users. Cisco expects WideField’s technology to support AI-driven security workflows and reasoning at scale, using identity telemetry to assemble session-level signals for deeper analysis and safer AI operations beyond security alone. A10 is positioning its TrojAI integration to capture long-term demand for secure AI infrastructure, emphasizing data sovereignty and continuous hardening. The direction is clear: siloed point products that only look at prompts or outputs will not survive once AI agents are woven into production-critical systems.
What This Means for Enterprises Adopting Agentic AI
For ordinary enterprise users, the practical impact is that AI security is becoming part of the infrastructure they already depend on, instead of a separate compliance box to tick. In Cisco’s world, Splunk’s Agentic SOC will use identity telemetry and intelligence from a wide range of sources to assemble session-level signals, giving analysts the context they need to decide whether an action belongs to a legitimate active session or a potentially malicious one. The same identity and session intelligence is expected to strengthen Cisco Data Fabric, giving customers more context to operate AI safely and at scale beyond security operations.
In A10’s ecosystem, AI threat mitigation will be native to the systems already protecting applications, APIs, and networks, mapping multi-modal agent execution traces and supervising sensitive operations. That unified landscape is designed to protect large-scale public sector and Fortune 50 corporate installations under the same umbrella that guards traffic and availability. The message to enterprises is pointed: if your AI agents are going to run core workflows, you will not be able to secure them with legacy, text-only filters or a scattered set of niche tools. You will need end-to-end AI infrastructure security embedded in the platforms that move your data, host your apps, and monitor your sessions. Vendors are now racing to own that foundation; buyers should judge them on how well they understand agent behavior, not how many dashboards they ship.
