AIR raises $50M to vet the tools AI agents use

AIR raised $50 million to check the skills and add-ons that AI agents pull into a company, and says it rejects 27% of what it evaluates. Founded six months ago by veterans of Israel's Unit 8200, the startup took $10 million from Sequoia first, then $40 million led by Greenoaks, per TechCrunch.

The product finds the agents already running inside a company, then continuously vets the tools, plugins and MCP servers those agents load, blocking the ones that fail. AIR has more than twenty customers, concentrated in financial services and pharma, two industries that can't afford an agent pulling in a broken or malicious tool undetected.

The number worth sitting with is the 27%. If a quarter of the agent add-ons a serious vetting process sees fail some basic bar, that's not a rounding error in tool quality. It's a sign that the ecosystem shipping MCP servers and agent skills has close to no quality floor, and most teams running agents have no visibility into which of their tools would fail the same check.

TechCrunch's report doesn't say what AIR's rejection criteria actually are, so it's hard to know if 27% means security flaws, broken functionality, or something looser like unclear permissions. That gap matters if you're trying to judge how alarming the number really is.

What it does confirm is investor appetite for agent infrastructure that isn't the model itself. AIR isn't building agents, it's building the layer that decides which tools those agents are allowed to touch, and two firms just bet $50 million that companies need that layer badly enough to pay for it before something breaks.