OCOPD AIR emerges from stealth with $50M for AI agent security
AIR $50M funding brings new attention to an emerging enterprise security problem: controlling the skills, plugins, MCP servers and other external components that give AI agents the ability to retrieve information and take actions. The startup has emerged from stealth with a $50 million round led by Sequoia Capital and Greenoaks, with Swish Ventures and Netz Capital also participating.
Founded by CEO Yair Saban and CTO Niv Hoffman, both veterans of Israel’s Unit 8200 intelligence corps, AIR is positioning its product as an inline firewall for AI agents. Its platform is designed to find agents operating across a corporate environment, identify AI tools being used outside IT approval processes, and inspect the components agents attempt to load or access.
The company says it can intervene at runtime, analyzing actions such as loading a skill or fetching internet content. Components are checked against a whitelist maintained by AIR; security teams can identify the agents and workflows dependent on a malicious, vulnerable or unapproved add-on, then revoke its use across the organization.
A supply-chain control layer for autonomous software
AIR’s core premise is that agentic systems expand the enterprise attack surface beyond conventional applications and code repositories. Agents can draw on external instructions, software packages, plugins and tool connections while operating, creating opportunities for compromised components or malicious content to influence their outputs and actions.
The platform covers a broad set of agent extensions, including:
- Skills and plugins used to extend agent capabilities
- MCP servers, which connect models to external tools and data sources
- Other add-ons and software components in an agent’s operational supply chain
- Unapproved AI services and personal-account usage that may sit outside formal IT oversight
AIR also plans to offer a marketplace of pre-vetted and certified add-ons. That could give enterprises a more controlled way to broaden the scope of agent deployments while reducing reliance on components that have not undergone security review.
Saban has argued that permissions alone are not the central risk in agentic systems. “We think that this is the source of the risk, not unnecessary permissions,” he said, referring to the information and tools agents encounter. “There are new challenges today and you need to know that everything it encounters is filtered.”
AIR says it rejects 27% of online agent extensions it evaluates
The startup says its whitelist is continuously updated because an extension that was initially approved can become unsafe later. A package downloaded by that extension may change, for example, or a developer account could be compromised.
According to AIR, its system currently filters out approximately 27% of the skills and add-ons it finds online. The company has also published research highlighting the potential scale of the exposure. In one study, AIR identified more than 17,800 public AI add-ons representing roughly 6.7 million installations that relied on untrusted external sources for instructions.
In another study, AIR said it found AI skills impersonating trusted brands, including Anthropic and OpenAI, in an effort to evade platform security reviews. One of those skills was capable of executing arbitrary code on enterprise systems, according to the company.
These findings are company research, rather than an independent measurement of the entire agent-extension market, but they illustrate why AI software supply-chain controls are becoming a priority for security teams.
Regulated industries are an early commercial target
AIR says it has more than 20 customers, about a quarter of which are large enterprises. Saban said demand has been strongest in heavily regulated sectors, particularly financial services and pharmaceutical companies.
Those industries face substantial pressure to establish controls over how sensitive data is accessed and how automated systems make or support decisions. AI agents add a complication: governance must cover not only the model or application an organization approves, but also the changing ecosystem of tools and data sources that the agent can call during operation.
Greenoaks partner Patrick Backhouse framed the shift as a fundamental change in enterprise security. “Software supply chains were never the most critical attack surface within an enterprise,” he said. “Now, with AI agents, they are vitally important.”
Funding signals a crowded but expanding AI governance market
The round places AIR in a growing market for products that secure, observe and govern enterprise AI. Existing cybersecurity vendors, cloud providers and AI-focused startups are all moving to address prompt injection, unsafe tool use, data leakage, shadow AI and model governance.
AIR’s differentiation is its focus on the agent-extension layer and real-time enforcement. That focus could be particularly relevant as companies move beyond chatbot pilots toward agents that interact with internal systems, browse external sources and execute multi-step workflows.
For buyers, the test will be whether AIR can provide reliable visibility across fragmented agent deployments without slowing them down or creating another difficult security-control plane. For the wider market, the funding reinforces a broader conclusion: as AI agents gain autonomy, the software and information surrounding them are becoming as important to secure as the underlying model.
Source
Original source: techcrunch.com