Salt Security and CrowdStrike are expanding their security integration to give enterprises greater visibility into how AI agents connect to APIs, Model Context Protocol (MCP) servers and internal systems. The integration links Salt’s Agentic API platform with CrowdStrike Falcon Foundry, Falcon Next-Gen SIEM and Falcon Firewall Management, extending protection into the runtime activity of increasingly autonomous enterprise AI systems.
3. Article
Salt and CrowdStrike target the security gap around AI agents
As enterprises move AI agents from experiments into production workflows, a new security problem is emerging: organizations may know which agents they have deployed without having a complete picture of what those agents can access or what they actually do.
Salt Security and CrowdStrike are targeting that gap with expanded integrations between Salt’s Agentic API security platform and the CrowdStrike Falcon ecosystem. The companies say the integration can help customers discover AI agents, the MCP servers and tools they use, the APIs those connections invoke and the permissions associated with each connection.
The announcement builds on a relationship that began in 2022, when CrowdStrike’s Falcon Fund invested in Salt Security. The companies subsequently introduced a certified Salt application for Falcon Foundry and integrated Salt with Falcon Next-Gen SIEM. The latest expansion extends that existing API-security relationship into agentic AI security.
The distinction is important because AI agents increasingly act as intermediaries between users and enterprise systems. Unlike a conventional chatbot that primarily generates an answer, an agent can call tools, retrieve information, update records or initiate other actions through APIs.
MCP has accelerated this model by providing a standardized way for AI applications to connect to external tools and data sources. That flexibility also creates another layer that security teams need to inventory and govern.
Gartner predicts that 40% of enterprise applications will incorporate task-specific AI agents by the end of 2026, up from less than 5% in 2025. The research firm has separately warned that organizations face growing governance challenges as AI agents proliferate.
That growth changes the security problem from protecting an application to tracking an agent’s entire path through the enterprise.
From API discovery to agent behavior
Under the expanded integration, organizations can deploy Salt through a certified Falcon Foundry application using existing Falcon sensors, without adding gateways or proxies, according to Salt. The technology can identify agents operating in an environment and map their connections to MCP servers, tools and APIs.
It can also expose connections that security teams may not have formally reviewed, including MCP servers accessible from the public internet and integrations that have entered an environment outside established security processes.
That inventory then becomes security telemetry. Salt’s findings can flow into Falcon Next-Gen SIEM, where agent inventory, posture information and behavioral detections can be correlated with endpoint, identity and cloud data already collected by CrowdStrike.
This is significant because agent security cannot necessarily be separated from the systems an agent interacts with. An agent might operate within its intended purpose while still presenting a security problem if it has excessive permissions, connects to an improperly secured API or reaches a system that its operator did not intend to expose.
CrowdStrike has been expanding its own AI security capabilities in the same direction. The company introduced Falcon Guardian in September as an AI Detection and Response capability designed to provide runtime protection for AI agents, including visibility into agent activity and the systems agents can access.
The Salt integration adds API and agent-connection context to that broader security environment.
When agent behavior deviates from an established baseline, customers can also trigger automated responses through Falcon Firewall Management, according to Salt. That creates a potential workflow spanning discovery, detection, correlation and response rather than leaving API or agent telemetry in a separate security console.
Agentic AI creates a new enterprise attack surface
The need for that visibility is becoming more pronounced as organizations deploy agents with increasingly broad permissions.
Gartner forecasts that the average Fortune 500 enterprise could have more than 150,000 AI agents in use by 2028, compared with fewer than 15 in 2025. Gartner describes the resulting proliferation as “agent sprawl,” with governance, data-loss and oversharing risks becoming more difficult to manage.
The numbers are forecasts rather than measurements of current deployments, but they illustrate why agent inventory is becoming an emerging security discipline.
Gartner has also predicted that by 2027, 40% of enterprises will demote or decommission autonomous AI agents because of governance failures discovered after production incidents. The research firm specifically points to the need to distinguish an agent’s level of autonomy from the scope of access it receives.
That distinction maps closely to the problem Salt and CrowdStrike are addressing. Knowing an agent is approved does not necessarily establish that every connection, tool and API available to it is appropriate.
The companies’ approach therefore extends traditional API security into AI agent security, while connecting those controls to an existing security operations platform. Salt’s technology focuses on the API and agent relationships, while Falcon provides a broader environment for correlating those findings with endpoint, identity, cloud and network telemetry.
The competitive landscape is expanding accordingly. Security vendors including Microsoft, Palo Alto Networks, CrowdStrike and specialized API-security companies are developing controls around AI applications and agents, while cloud and AI platforms are adding governance mechanisms of their own.
For enterprise security teams, the emerging requirement is less about simply blocking AI use and more about establishing a reliable chain of visibility: which agents exist, what they can access, which tools they invoke, what APIs they call and whether their behavior remains within the intended boundaries.
Salt and CrowdStrike’s expanded integration is aimed at that chain. As enterprise AI moves toward systems capable of taking action rather than simply generating content, the security architecture around those actions is becoming part of the AI infrastructure itself.
4. Market Landscape
The enterprise AI security market is shifting from protecting models and prompts toward securing agents, tools, APIs, identities and runtime actions. Gartner’s forecasts point to rapid growth in task-specific agents alongside significant governance challenges.
CrowdStrike is building agent security into its broader Falcon platform, while Salt brings specialized API and agentic-security capabilities into that environment. The integration reflects a broader convergence between AI security, API security, SIEM, identity security and automated response.
The key architectural challenge is maintaining visibility across the full agentic path. An AI model may be secure while the API, MCP server or downstream system it accesses remains vulnerable. Security platforms therefore increasingly need to correlate AI behavior with the infrastructure and permissions surrounding it.
5. Top Insights
- Salt and CrowdStrike are expanding their existing API-security partnership to provide visibility into enterprise AI agents, MCP connections and downstream APIs.
- The integration connects agent inventories and behavioral findings with endpoint, identity and cloud telemetry inside Falcon Next-Gen SIEM.
- Gartner predicts 40% of enterprise applications will feature task-specific AI agents by the end of 2026.
- Automated response through Falcon Firewall Management extends the integration beyond visibility into potential runtime containment.
- The emerging agentic security model focuses on what AI systems can access and do, not only what they generate.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
