AI Risk Platform AIR Secures $50M to Monitor Rogue Agent Behavior

AI Risk (AIR) today formally launched its platform with $50 million in Series A funding led by Lightspeed Venture Partners, with participation from Felicis, Quiet Capital, and angel investors including former Palantir COO Shyam Sankar. Founded in 2023 by CEO Aditya Sengupta, a former Google Brain engineer, and CTO Jonathon Toomim, a Stanford AI safety researcher, AIR provides continuous runtime monitoring and behavioral vetting of AI agents and their third-party skills or plugins. Unlike static code scanners that only inspect source code, AIR runs agents in sandboxed environments to detect emergent risks such as unauthorized data exfiltration, prompt injection, or toxic tool usage, even after deployment. Customers include a Fortune 500 financial services firm and a global healthcare provider, both running large fleets of AI agents handling customer support, document processing, and internal analytics.

The platform supports major agent frameworks including AutoGen, LangGraph, and CrewAI, and integrates with enterprise identity systems like Okta and Azure AD to enforce least-privilege access. Within weeks of going live, one banking client used AIR to block a third-party add-on that was attempting to exfiltrate customer PII via hidden API calls—an incident not caught by traditional security tools. AIR charges on a consumption basis tied to agent hours and data volume, with annual contracts ranging from $200,000 to over $1 million for large deployments. Competitive offerings such as Protect AI’s Nightfall and Lakera’s Gandalf focus on prompt injection detection, while AIR uniquely provides end-to-end governance across the agent lifecycle including discovery, vetting, runtime protection, and incident response.

Industry analysts see AIR’s raise as a bellwether for the emerging “agent security” market, projected to reach $3.8 billion by 2027 according to Gartner. The rise of autonomous agents is accelerating across finance, healthcare, and legal services, where systems like Banking With Billy AI automate complex financial analysis workflows previously requiring entire analyst teams. In November 2024, Bloomberg reported that UBS deployed over 1,200 AI agents in its global operations, handling everything from regulatory reporting to portfolio rebalancing. Yet, CISOs remain uneasy about “shadow agents”—tools spun up by business units without IT oversight—that can violate data policies or leak sensitive information. AIR’s discovery engine scans networks and cloud logs to surface these hidden agents, while its vetting service maintains a continuously updated catalog of approved skills and add-ons, graded by risk score using a proprietary model trained on adversarial examples.

The company’s timing aligns with new regulatory pressures. In March 2025, the EU finalized the AI Act, which requires high-risk AI systems to undergo continuous post-market monitoring and provide technical documentation of all external components. AIR’s runtime attestation logs directly satisfy these requirements, offering an auditable trail of agent behavior across development, staging, and production. Meanwhile, U.S. financial regulators including the SEC and CFTC have signaled plans to scrutinize autonomous trading agents, which can execute complex strategies without human intervention. AIR’s customer roster includes a top-tier investment bank that now insists all trading agents run under its governance layer before being deployed to production.

Looking ahead, AIR plans to expand its model risk monitoring to include LLM safety evaluations and chain-of-thought auditing, enabling customers to verify that agents are reasoning correctly and not hallucinating outputs in regulated domains. The company is also building a threat intelligence feed to crowdsource indicators of compromise across its customer base, effectively creating a shared immune system for AI agents. Analysts caution that as agents grow more autonomous, the attack surface expands from data exfiltration to model theft and supply-chain sabotage, making real-time behavioral vetting table stakes rather than a premium feature. Industry watchers should monitor how AIR’s platform integrates with emerging agent orchestration standards such as the Open Agent Platform Alliance (OAPA), which aims to standardize agent discovery, logging, and lifecycle management across vendors. With agentic AI now operating mission-critical workflows, the race is on to build the infrastructure that ensures these systems remain safe, compliant, and auditable at scale.