As enterprises increasingly grant AI agents deeper access to their systems, a new software supply chain is emerging—one built around the tools agents use: skills, plugins, MCP servers, and add-ons that connect them to the internet.
AI security startup AIR is betting that companies will need robust oversight of this supply chain. Emerging from stealth today, AIR has secured $50 million in two seed rounds to bring its platform to market, addressing a critical gap in how AI agents are deployed securely.
What AIR Does
Founded by Yair Saban (CEO) and Niv Hoffman (CTO)—veterans of Israel's Unit 8200 intelligence corps, where they specialized in offensive cybersecurity—AIR offers a platform that discovers AI agents running across a company's environment. It continuously vets the skills, tools, and components those agents use, blocking interactions with software or external sources that fail security criteria. The company also operates a marketplace of vetted add-ons and skills for AI agents.
This comes at a pivotal time. By 2026, Gartner predicts that 40% of enterprise AI deployments will include agentic systems, yet security standards for agent tooling remain nascent. AIR aims to be the much-needed enforcement layer.
Funding Details
The two funding rounds closed just weeks apart, Saban told TechCrunch. The first raised $10 million, led by Sequoia, and the second $40 million, led by Greenoaks. Additional investors include Swish, Netz, and notable angels such as Zach Frankel (president of Cognition), Yinon Costica (co-founder of Wiz), Ofir Erlich (co-founder of Eon), Anne Neuberger, Omer Adam, and Varun Anand (co-founder of Clay).
The Driver Analogy: A Cautionary Tale
Saban draws a parallel to the early 2000s, when installing a device driver didn't require a signature. Today, signed drivers are standard because they load code directly into the kernel. "You don't have that with skills or plugins or MCPs," Saban noted. "It's a shame, because it's the same mechanism, the same lesson—but we haven't learned it yet."
The core risk, he argues, is that as AI agents operate autonomously across databases, enterprise systems, and the internet, attackers may poison the content an agent consumes rather than attacking the agent directly. This indirect attack vector is increasingly common in 2026 threat landscapes.
How AIR Works
The platform's approach is threefold:
- Visibility: Discovers active agents and flags unauthorized AI tool usage, including personal accounts.
- Enforcement: Hooks into agents to intercept and analyze actions—such as loading a skill or fetching web content—against security policies.
- Vetting: Checks tools, add-ons, and software against a whitelist maintained by AIR, which continuously evaluates openly available skills and add-ons for changes or malicious behavior, recognizing that an approved skill can become risky if its underlying package changes or its developer's account is compromised.
Currently, AIR filters out roughly 27% of the add-ons and skills it encounters online, Saban said.
Early Traction and Target Industries
The startup claims more than 20 customers, about a quarter of which are large enterprises. Demand has been strongest in highly regulated sectors, particularly financial services and pharmaceuticals, where governance and security are paramount.
As AI agent adoption accelerates, AIR's platform could become essential infrastructure for enterprises aiming to balance innovation with safety in an increasingly connected digital ecosystem.
via TechCrunch