AI agent security M&A map forms around control points, not one category

As AI agents browse the web, write code, access files, trigger APIs and interact with internal systems, they are becoming a new class of enterprise identity. Unlike traditional users or service accounts, these identities are not passive: they can move between systems, invoke tools and make decisions, which makes controlling them more complex. Companies will need to know which agent accessed what information, which systems it connected to, and whether its actions were authorized.

The article suggests this market will not develop as one broad category called 'AI security.' Instead, the real opportunity will be around specific control points — one company may protect agent identity, another may control the data an agent can access, while others may focus on prompts, MCP servers, plug-ins, traffic or auditability. Existing activity supports this view: Kiteworks acquired Bonfy.AI for real-time data classification and policy enforcement, while Huskeys raised a $27 million Series A led by Blackstone to secure complex internet traffic, including traffic generated by autonomous systems.

For entrepreneurs, this means 'AI security' may already be too broad a positioning. The more important question is what exactly the company controls. The stakes hinge on whether these control points consolidate into broader platforms or remain standalone tools, and on whether identity providers, data-security vendors, cybersecurity platforms, cloud companies and enterprise software vendors embed agent-security capabilities directly into their products. How that plays out will determine which startups become acquisition targets and which become features of larger platforms.