The market moved to the agent
This turned into a budget line fast. In Latio's survey, the share of security teams with a dedicated AI security budget rose from 8% to 37% in twelve months.
The concern has moved too. In Latio's survey, 45% of respondents named endpoint agents as their primary AI security concern, ahead of first-party agents (23%), browser applications (22%) and hosted agents (10%). Latio calls the endpoint the second era of AI security, and lists governance and supply chain malware prevention for skills and MCP servers as one of the three outcomes that era has to deliver.
The surface inside the agent
Once the agent is the thing you're securing, the next question is what the agent consumes. Agents on employee laptops run skills, MCP servers, plugins and subagents. Each add-on extends what the agent can do. Each one is also a set of instructions the agent will follow.
Traditional supply chain tools check code. Many agent add-ons are written in natural language, and they can pull more instructions from places nobody reviewed. We've seen this in our own research:
- We scanned 142,836 live skills and found thousands depending on external resources we classified as untrusted: domains, GitHub repositories, code packages and hosting infrastructure.
- Across more than 17,800 public add-ons with 6.7 million installations, AIR filters out roughly 27%.
- An official OpenAI skill for deploying apps on Render told those apps to send authentication tokens to a domain nobody owned. We registered it and confirmed live production apps were sending real tokens.
What Latio said about AIR
"A key differentiator is AIR's filtering of every agent operation, such as loading a skill or running a webfetch, to protect before an incident occurs. Coupled with their deep understanding of permissions, this makes AIR able to spot the moment a session is hijacked."
— James Berthoty, Charrah Hardamon, and the Latio team, 2026 AI Security Market Report
AIR covers that surface through four products that work as one. AIR Control discovers every agent and what it has installed. AIR Filter vets every add-on before install and on every load. AIR Defend secures agent actions at runtime. AIR Marketplace gives the enterprise one trusted source of approved add-ons.
How to use the report
- Define the surface you care about before comparing vendors: browser use, endpoint agents, first-party agents, or what those agents install.
- For each vendor, ask where its control sits and what it can see from there.
- Pick the partner whose direction matches where your agent program is going, not only today's feature list.


%20(1).png)


.png)


%20(1).png)