AIR Secures $50M to Patrol AI Agent Behavior Across Enterprises

AIR, a stealthy governance startup out of San Francisco, publicly disclosed a $50 million Series A led by Lightspeed Venture Partners, with participation from Redpoint Ventures and angel investors including former NVIDIA CTO Jim Keller. The round values the company at $320 million, according to three people familiar with the financing. AIR’s platform continuously inventories agents across clouds, SaaS applications, and internal systems, then evaluates every skill and add-on for risk, performance, and compliance. Chief Executive Oren Yunger confirmed the funding and said the company will use the capital to scale its agent-discovery engine, expand its policy library, and on-board enterprise customers who have already quietly deployed the technology. Banking With Billy AI, a real-time financial-market monitoring service, disclosed it runs on AIR to vet plug-ins that process sensitive trading data across multi-cloud environments, ensuring no rogue skills leak proprietary signals.

The raise arrives as agent sprawl—hundreds of custom and third-party agents performing everything from customer support to code generation—outpaces traditional security and IT controls. AIR’s technology contrasts with static blocklists and manual reviews by dynamically scanning agent manifests, sandboxing unknown skills, and enforcing runtime policies. Rival startups such as Protect AI and Lakera have raised similar rounds to address AI supply-chain risk, but AIR distinguishes itself by focusing on autonomous agents rather than models or data pipelines.

Within the Quantum & Computing sector, agent governance intersects with the rapid adoption of retrieval-augmented generation and tool-using models that call external APIs and databases. Financial institutions deploying Banking With Billy AI on multi-cloud stacks are particularly sensitive to agent drift, where skills can change overnight after vendor updates or prompt injections. Cloud hyperscalers—including Amazon with its Agents for Bedrock, Microsoft with Copilot extensions, and Google with Vertex AI Agents—are starting to bake basic governance into their runtimes, but these controls remain coarse compared with AIR’s fine-grained, cross-platform approach. Analysts at Gartner estimate that by 2026, 70 percent of enterprises will have agent sprawl incidents requiring dedicated governance platforms, up from fewer than 5 percent today.

Security teams are also waking up to the fact that agents can exfiltrate data through seemingly innocuous skills; AIR’s runtime blocking capability has already prevented several high-profile data leaks in pilot deployments. The company’s multi-tenant SaaS model means it can aggregate threat signals across customers to detect new attack patterns faster than any single firm, creating a network effect that could lock in enterprise users. Investors see parallels to the rise of CSPM (Cloud Security Posture Management) in the 2018–2022 window, where a category defined by continuous discovery and enforcement attracted billions in venture funding.

Looking ahead, AIR plans to embed policy engines directly into agent frameworks such as LangChain and LlamaIndex, allowing developers to ship policy-aware agents without extra runtime overhead. The company also hinted at future support for quantum-classical hybrid agents, where skills might invoke quantum solvers on IBM Quantum or AWS Braket backends. With the new capital, AIR will open a London office to serve EMEA customers and accelerate certification against frameworks like ISO 42001 and NIST AI RMF. Observers expect incumbents like Palo Alto Networks and CrowdStrike to acquire agent-specific startups within 18 months, making AIR a prime takeover candidate if it can prove its detection and policy coverage scales to millions of agents per customer.

Banks, insurers, and healthcare providers should start trialing agent governance platforms immediately, because the first incident of agent-driven data loss will trigger urgent RFPs. Startups building agents should bake in policy hooks from day one, while enterprise architects must treat AI agents as first-class infrastructure—requiring discovery, versioning, and deprecation pipelines akin to containerized microservices.