Why Enterprise AI Security Funding Is Exploding

The numbers coming out of the AI security market look completely disconnected from how quiet the headlines are.

Gartner projects spending on securing AI tools will jump 83% this year to $2.83 billion, heading toward nearly $5 billion by 2027. HiddenLayer just closed a $100M Series B after claiming a 10x ARR surge. Noma and Zenity have both raised nine-figure rounds of their own.

What struck me about this money wave is where it is going. When security startups were raising $50 million Series A rounds three years ago, the pitch was theoretical protection against prompt injection and model extraction. Nobody really had evidence of agentic exploits running at scale in production.

Now the security stack is morphing into classic endpoint detection and response, tuned specifically for autonomous agent runtimes. Companies are not just scanning static model weights. They are monitoring execution loops, auditing open-weight file frameworks for hidden payloads, and trying to stop agents from executing malicious tool calls.

The real tension here is whether standalone AI security platforms can survive alongside bundled cloud features. Large cybersecurity incumbents like Cisco, Palo Alto Networks, and Check Point are buying up AI security startups to plug into their existing platforms, while model builders add governance directly into their infrastructure.

I keep coming back to the open-weight risk vector. As more enterprises run self-hosted models, downloading open weights off public hubs becomes a direct software supply chain vulnerability. Parsing fifty different model formats to verify that a file actually contains what it claims to be sounds unglamorous, but that is precisely where production systems break down.

Protecting the agent harness in real time is where the money is flowing.