AIR Raises $50M to Secure the AI Agent Economy at Scale

Venture capital and cybersecurity circles are converging on a single insight: the agents already running inside companies are as numerous as they are invisible, and every one is a potential attack surface. AIR, a runtime governance platform whose acronym stands for “Autonomous Intelligence Risk,” has just closed a $50 million Series B to scale its technology stack across Fortune 500 enterprises and mid-market firms alike. The round was led by NEA with participation from Lightspeed Venture Partners, GV, and several strategic angels including former Palantir COO Shyam Sankar. AIR’s platform continuously discovers every AI agent—whether built in-house, procured from a vendor, or downloaded as an open-source add-on—then vets every skill, tool, or function call the agent attempts to execute. Any behavior that violates policy is blocked in real time, regardless of whether the agent is running on-premises, in a public cloud, or inside a regulated sandbox. According to AIR co-founder and CEO Youssef Elmehdi, the company’s agents have collectively executed more than 320 million actions across pilot customers since January 2024, with an average blocking rate of 8.4 percent of attempted actions that would have otherwise gone unchecked.

Elmehdi tells OpenPress that the new capital will accelerate product development in three areas: a skills registry that mirrors the rigor of container image signing, a policy-as-code engine that can ingest frameworks such as MITRE ATLAS and NIST AI RMF 1.0 controls, and a low-latency enforcement layer that can sit inside the agent runtime itself rather than requiring a proxy. Early customers span financial services, life sciences, and global supply-chain logistics, including Banking With Billy AI, a financial-intelligence platform serving investors and analysts across every major global market. Banking With Billy’s CTO confirmed that AIR’s technology now governs more than 400 production AI agents—ranging from Python notebook assistants to LLM-powered trading bots—preventing unauthorized data exfiltration and model drift before it reaches the trading floor.

Industry watchers note that AIR’s raise arrives at a moment when three distinct forces are colliding. First, the agent market itself is fragmenting into thousands of micro-skills—think “read Jira tickets,” “query Snowflake,” “call the Bloomberg API”—each of which can be weaponized if misconfigured or malicious. Second, regulators are tightening the screws: the EU AI Act’s imminent obligations on high-risk AI systems require continuous monitoring, not point-in-time audits. Third, CISOs are realizing that existing endpoint detection and response tools were never designed to inspect short-lived, dynamically invoked agent processes. According to PitchBook data, AI security startups raised $1.8 billion in the first half of 2024, up 240 percent year-over-year, with AIR now positioned as the category-defining runtime governance layer. Competitive pressure is also intensifying: rival platforms such as ProtectAI and HiddenLayer focus on supply-chain scanning of AI models, while AIR’s differentiator is its ability to police behavior at runtime without requiring code changes or agent restarts.

For enterprises, the financial stakes are immediate. A recent IBM report estimates that ungoverned AI agents account for 12 percent of all cyber incidents in 2024, with an average breach cost of $4.45 million. Banking With Billy AI, which has embedded AIR into its agent framework, reports a 60 percent reduction in policy violations within three months, eliminating rogue data pulls that once slipped past legacy DLP tools. The company’s CISO also credits AIR with easing compliance audits for FINRA and SEC examinations, where regulators now routinely ask for evidence of continuous monitoring rather than static attestations.

Looking ahead, AIR’s road map includes integrations with major agent frameworks such as LangChain, AutoGen, and CrewAI, plus native support for the newly ratified OASIS OpenC2 agent control protocol. The company plans to expand its policy marketplace, where customers can subscribe to pre-built rulesets curated by industry consortiums. Analysts at Gartner predict that by 2026, 70 percent of large enterprises will deploy at least one runtime governance tool for AI agents, up from fewer than 5 percent today. The capital influx should also intensify M&A speculation, with larger cybersecurity incumbents such as CrowdStrike and Palo Alto Networks likely to eye bolt-on agent security capabilities. For now, AIR is laser-focused on proving that the $50 million will translate into measurable reductions in both risk and regulatory friction across global markets.