The September 1 AI-security tape did not only print platform M&A. It printed a seed-stage category bet with grown-up capital.
AIR emerged from stealth and announced $50 million raised across two seed rounds that, per CEO Yair Saban speaking to TechCrunch, closed within weeks of each other: $10 million led by Sequoia Capital, then $40 million led by Greenoaks. Swish, Netz, and a roster of angels — including Zach Frankel (Cognition), Yinon Costica (Wiz), Ofir Ehrlich (Eon), Anne Neuberger, Omer Adam, and Varun Anand (Clay) — participated. Ctech and AIR’s newswire release match the $50 million headline and the Sequoia / Greenoaks lead structure. No post-money valuation appears on any of those sources. CyberMerge will not invent one.
Founders Yair Saban (CEO) and Niv Hoffman (CTO) are described as Unit 8200 veterans with offensive cybersecurity backgrounds; the company dates to February 2026. TechCrunch puts headcount around 40 and says capital will go to researchers plus U.S. and Europe go-to-market. AIR claims more than 20 customers, with roughly a quarter described as large enterprises and strongest demand in regulated financial services and pharma. Those are company claims, not audited filings.
The product thesis is specific. As agents pick up skills, plugins, MCP servers, and add-ons — a messy software supply chain that looks more like unsigned drivers than an app store — AIR wants to discover which agents are running, continuously vet what they load, and block anything that fails a whitelist. Saban’s analogy in TechCrunch is kernel drivers before code signing. The company says its evaluation pipeline currently filters out about 27% of the add-ons and skills it finds online. Sequoia partner Bogomil Balkansky framed it as continuous re-verification infrastructure, not a better one-time scanner.
Competition is not theoretical. TechCrunch itself names Noma Security, Zenity, Astrix, and Operant AI as overlapping. Zenity just raised a $125 million Series C in August. That is the point of the AIR print: the agent-skill / MCP governance layer is already a funded category fight, not a blank whiteboard. AIR’s differentiation claim is the hard pipeline that re-checks changing packages and compromised publisher accounts in real time. Buyers will test whether that is a moat or a marketing sentence.
Same-day context matters. Palo Alto bought Console to agentify Cortex. CrowdStrike launched Falcon Guardian for runtime AIDR at the endpoint. AIR is selling an independent firewall for the skill supply chain those platforms will eventually claim as a SKU. Early capital from Sequoia and Greenoaks buys time. It does not buy category ownership.
Underwrite what is on the page: $50 million, named leads, a supply-chain framing for agent tools, customer and headcount claims from the company, and a blank valuation. Ask for the mark before you underwrite the “firewall for agents” slogan. Capital without a price is still a signal. It is not a completed argument.