Air Security, which builds a security service for extensions and other tools installed on AI agents, emerges from stealth with $50M led by Sequoia and Greenoaks
The Israeli startup emerged from stealth with research showing thousands of AI add-ons rely on untrusted sources …
Context & Ripple Effects
Air Security enters an expanding Israeli security cohort focused on risks created by AI-enabled work: Bloom Security launched to monitor AI agents and extensions, while Ent Security has positioned around interpreting user and agent intent. Air narrows its focus to the tools and extensions an agent is allowed to use.
Its research on add-ons drawing on untrusted sources makes the agent toolchain a distinct control point, rather than merely another endpoint or cloud-access problem. The $50M round gives that thesis substantial backing as Act Security targets cloud access-surface reduction from an adjacent layer.
First-order effects
- Enterprises using AI agents gain a dedicated option for vetting the extensions and tools those agents can invoke, addressing exposure from add-ons that rely on untrusted sources.
- Air Security has funding from Sequoia and Greenoaks to build and sell that agent-tool security layer after emerging from stealth.
Second-order effects
- Bloom Security and Ent Security face a more segmented buyer conversation: monitoring agent behavior and reading intent must be differentiated from approving the agent's underlying toolchain.
- Enterprise security teams evaluating AI-agent deployments may treat extensions and tool sources as a separate review and policy category alongside endpoint, workspace, and cloud controls.
Third-order effects
- If agent deployments continue to rely on third-party add-ons, security architecture is likely to add dedicated trust controls around what agents are permitted to call, widening the agentic-security stack beyond endpoint protection.
- The cluster of well-funded Israeli entrants—from Frame Security's AI social-engineering defense to Air's toolchain focus—points to AI security splitting into specialized control layers rather than one general-purpose category.
The trend: AI security is fragmenting around the distinct attack surfaces created by agents, including the tools, extensions, identities, and actions that connect them to enterprise systems.