AIR secures $50M to police rogue AI agents with real-time vetting

AIR, a stealthy Palo Alto-based startup, disclosed today a $50 million Series A led by Lightspeed Venture Partners with participation from GV, Felicis, and several industry angels. The round values AIR at $280 million post-money and comes barely twelve months after the company emerged from stealth with a seed round of $12 million. Co-founders CEO Ravi Iyer and CTO Sophie Laurent unveiled their platform in March 2024, positioning it as the first continuous compliance engine built expressly for AI agents—software entities that can plan, use tools, and even hire other agents without human intervention.

Iyer described the product as “a security and governance layer that sits between the orchestration platform and the agent’s runtime environment,” scanning every skill, toolkit, and third-party add-on that an agent might invoke. AIR’s runtime monitor intercepts calls to external APIs, checks against a growing threat intelligence graph, and can quarantine or block any behavior that deviates from policy. Early customers such as banking giant BillyCorp have used the platform to govern Billy AI, the bank’s proprietary agent that automates complex financial analysis workflows previously requiring entire analyst teams. In one deployment, AIR flagged an unapproved add-on that was scraping sensitive deal data from internal systems—an incident caught within minutes of the attempt.

The company’s technical edge rests on three proprietary components: a discovery service that enumerates all agents running across cloud, data center, and edge nodes; a policy engine that encodes enterprise risk tolerances; and a continuous vetting pipeline that pulls data from public advisories, vendor APIs, and AIR’s own honeypot network. In a controlled benchmark with a Fortune 500 retailer, AIR detected 14 previously unknown agent behaviors in a single week, including an experimental add-on that was silently exfiltrating inventory feeds to a third-party logistics provider. Iyer claims the platform now covers more than 80 percent of the top twenty enterprise agent frameworks, including LangGraph, AutoGen, and CrewAI.

Funding news coincides with a sharp uptick in regulatory scrutiny. The U.S. Treasury’s Financial Crimes Enforcement Network issued draft guidance in April 2025 requiring financial institutions to “maintain real-time monitoring of autonomous agents with access to material non-public information.” AIR’s platform directly maps to that requirement, providing an audit trail and immutable logs that can be exported to regulators. Lightspeed general partner Priya Kapoor said the firm backed AIR because “governance is the last unsolved problem before enterprises can safely deploy thousands of agents at scale.” She expects the governance software market to exceed $4 billion by 2028 as CIOs seek tools that can keep pace with the velocity of AI agent adoption.

Industry Impact and Significance

Beyond BillyCorp, AIR counts two other Fortune 100 banks and a global shipping conglomerate among its paid pilots, each targeting governance for AI agents that handle trade finance, FX hedging, and cargo routing. The shipping firm’s deployment alone involves more than 3,000 agents coordinating across ports, carriers, and customs brokers—a scale that would overwhelm traditional SOC tooling. Competitors such as Torq and Styra, historically focused on workflow orchestration and policy-as-code respectively, are now adding agent-specific modules, but none yet offer the runtime discovery and continuous vetting that AIR delivers. Analysts at Gartner predict that by 2026, 60 percent of large enterprises will require agent governance platforms, creating a $2.4 billion revenue opportunity just for the continuous vetting segment.

Financially, the Series A signals investor appetite for infrastructure that reduces downside risk rather than chasing top-line growth. GV partner Mark Chen noted that “companies are willing to pay premium prices for software that can prevent a single rogue agent from leaking customer PII or triggering a systemic compliance violation.” The funding will accelerate hiring—especially in threat research and policy engineering—and expand AIR’s threat intelligence network into Europe and Asia, where local regulatory regimes are tightening around AI usage. AIR also plans to open-source a lightweight SDK later this year, aiming to crowdsource detection rules while maintaining a core commercial product for enterprise scale.

The Bigger Picture

AIR’s emergence reflects a fundamental shift from model-level governance to agent-level governance. Early AI safety efforts concentrated on model cards, bias audits, and content filtering, but the rise of multi-agent systems that can recruit tools, spin up sub-agents, and evolve their own workflows demands a new architectural layer. In many ways, AIR is filling the same role that endpoint detection and response tools did for traditional IT estates in the mid-2010s, but applied to a much more dynamic and interconnected threat surface.

The broader trend is mirrored in parallel initiatives: the Linux Foundation’s Open Policy Agent project has added support for agent policies, while the NIST AI Risk Management Framework now includes explicit controls for “agent autonomy and tool use.” At the same time, geopolitical blocs are racing to define standards—AIR has already aligned its policy engine with the EU AI Act’s high-risk classification, ensuring continuity for multinational clients. If adoption curves hold, agent governance could become a mandatory line item in every enterprise IT budget, much like SIEM or DLP today.

Expert Analysis

According to Ravi Iyer, the next inflection point will be “the moment when the average Fortune 500 company runs more compute hours in AI agents than in traditional microservices.” To reach that threshold safely, enterprises will need continuous runtime vetting embedded into their CI/CD pipelines, not bolted on as an afterthought. In the next eighteen months, watch for AIR to integrate directly with agent orchestrators such as LangGraph Cloud and AutoGen Studio, enabling policy enforcement at the point of agent creation rather than retroactively. Meanwhile, incumbents like Palo Alto Networks and CrowdStrike are likely to acquire or partner with niche governance players, consolidating the market into a handful of platform-level solutions. For CIOs, the message is clear: the governance budget is no longer optional—it is the price of admission for the agent economy.