Analysis
AIR, a startup that discovers and continuously vets the skills and add-ons AI agents use inside a company, raised $50 million across two seed rounds and emerged from stealth this week, TechCrunch reported. Swish, Netz and a group of angels including Cognition president Zach Frankel, Wiz co-founder Yinon Costica and Clay co-founder Varun Anand also participated.
- First seed -- $10 million, led by Sequoia Capital
- Second seed -- $40 million, led by Greenoaks Capital
- Combined total -- $50 million across both rounds
What AIR Actually Does
Founded by CEO Yair Saban and CTO Niv Hoffman, both veterans of Israel's Unit 8200 intelligence corps, AIR builds a platform that finds AI agents already running inside an organization -- often deployed by individual teams without central IT's knowledge -- and continuously monitors and vets the external skills, plugins and tools those agents pull in. It maintains a whitelist of approved add-ons and a marketplace of vetted components, blocking agents from interacting with unapproved external tools or data sources. That's a direct response to the same problem Pulse has covered from multiple angles this week: METR disclosed that an attacker stole an API key and burned $600,000 in model credits over three weeks undetected, and OpenAI and Hugging Face have both dealt with agent-security incidents recently. The operational surface around AI agents -- what tools they can call, what data they can touch, what happens when an agent's permissions get compromised -- has outpaced most enterprises' existing security tooling.
AIR already has more than 20 customers, roughly a quarter of them large enterprises, with the strongest early demand coming from financial services and pharmaceutical companies -- two industries where regulatory compliance requirements make "we don't fully know what our AI agents can access" an unacceptable answer rather than a manageable risk.
A Crowded, Fast-Forming Category
AIR's competitors include Noma Security, Zenity -- which raised a $125 million Series C in August, per the same reporting -- Astrix Security and Operant AI, all building some version of visibility and governance for the AI-agent layer of the enterprise stack. That's a lot of well-funded competition for a company that just emerged from stealth, and the category is forming fast enough that differentiation will likely come down to integration breadth (how many agent frameworks and enterprise systems a platform actually covers) and speed of threat detection, rather than any single company owning a defensible technical moat early.
The unusually large gap between AIR's first and second seed rounds -- $10 million to $40 million -- suggests the company found product-market fit fast enough that investors moved before a priced Series A, a pattern showing up across AI-security specifically as venture capital chases the same category CrowdStrike, Palo Alto Networks and Lasso Security are all racing to own from different entry points.
What's untested: whether AIR's whitelist-and-marketplace model scales to the sprawl of agent frameworks now shipping from OpenAI, Anthropic, Google and dozens of open-source projects, each with different plugin and tool-calling architectures. A governance layer that can't keep pace with how fast the underlying agent ecosystem is fragmenting becomes a compliance checkbox rather than a real security control.
