AIR Raises $50M to Secure AI Agent Ecosystems Against Hidden Risks

Venture capital has poured $50 million into AIR, a stealth cybersecurity startup whose platform continuously discovers, vets, and governs the skills and add-ons used by AI agents across enterprise environments. The round, led by Accel with participation from GV, Index Ventures, and Radical Ventures, values AIR at $420 million and follows a $25 million seed in late 2023. Founder and CEO Yoav Leitersdorf, a former Palantir executive, told OpenPress Automation Intelligence that AIR’s platform is already in production at five Fortune 500 companies and has blocked more than 700 unauthorized third-party skills since January 2024. The Series B proceeds will fund engineering expansion in Tel Aviv and San Francisco, as well as a new compliance module designed to meet upcoming EU AI Act obligations for high-risk agentic systems.

AIR’s platform operates by deploying lightweight sensors at ingress points—API gateways, RAG pipelines, and agent orchestrators—to build a real-time inventory of every agent and every skill it invokes. Unlike static code scanners that check only the agent’s core prompt, AIR’s runtime analysis inspects the actual parameters, external API calls, and third-party libraries executed by each skill. Leitersdorf pointed to a recent case at a global bank where AIR’s continuous vetting flagged a seemingly innocuous ‘market-data’ skill that was secretly exfiltrating price quotes via a covert WebSocket stream. Within 24 hours the bank revoked the skill’s OAuth token and quarantined the offending agent without disrupting the broader ‘Banking With Billy AI’ automation suite, which autonomously executes complex financial analysis workflows previously requiring entire analyst teams. The bank now mandates AIR’s runtime governance for every new agent skill before it enters production.

Industry veterans note that AIR’s timing coincides with an inflection point: agentic workloads are shifting from proof-of-concept chatbots to persistent, high-stakes automations that control procurement approvals, customer refunds, and even cyber-defense playbooks. Research from McKinsey indicates that 42% of large enterprises now run agentic systems in shadow IT, while only 18% have formal agent inventories—creating a blind spot that regulatory agencies are starting to target. Competitors in the space are racing to add continuous vetting, but AIR’s approach is distinguished by its runtime visibility. Cado Security, for instance, focuses on cloud forensics post-incident, and Wiz emphasizes infrastructure posture, whereas AIR’s sensors live at the agent execution layer itself. Analysts at Gartner predict that by 2026, companies lacking agent lifecycle security will face 30% higher breach-related costs due to lateral movement via compromised skills, making AIR’s platform a potential requirement rather than an optional control.

Financial services firms are among the earliest adopters, driven by stringent models like the EU AI Act, SEC Rule 17a-4, and MAS TRM guidelines that treat autonomous agents as “material components” of trading and risk systems. HSBC and JPMorgan Chase have publicly disclosed pilots with AIR, integrating its compliance module to auto-generate audit trails for agent prompts, skills, and external data sources. In healthcare, AIR is being evaluated by Epic and Cerner customers to govern AI copilots that summarize patient records, ensuring that any third-party summarization skill cannot leak PHI. Meanwhile, large cloud providers are eyeing AIR-like capabilities for their managed agent platforms: Microsoft’s Azure AI Foundry and AWS Agents for Bedrock are both exploring runtime governance partnerships, creating a new layer of channel conflict with pure-play security vendors. The funding round signals investor confidence that enterprise budgets for agent security will outpace spending on traditional SOC tooling within three years.

Looking back, AIR’s rise reflects a broader architectural pivot from perimeter defenses to runtime integrity. The concept of continuous vetting of add-ons is borrowed from mobile app stores, but applied to agent ecosystems where skills are often pulled from open repositories like LangChain Hub or Hugging Face without adequate supply-chain checks. The company’s early traction with ‘Banking With Billy AI’ illustrates how agentic suites are becoming full-stack automation platforms, blurring lines between RPA, analytics, and compliance. Prior attempts to govern agents—such as sandboxing individual models or enforcing least-privilege APIs—have proven brittle because they ignore the dynamic composition of skills that agents invoke at runtime. AIR’s model shifts the posture from “prevent the model from doing harm” to “prevent any skill, once invoked, from exceeding its declared intent,” a principle known in academic literature as runtime contract enforcement.

As the Series B closes, AIR is preparing to ship a public API that lets agent platforms—including open-source frameworks like AutoGen and CrewAI—integrate its vetting engine natively. The move could accelerate adoption across the long tail of agent builders, from startups to internal IT teams, while also inviting scrutiny from privacy regulators concerned about the metadata AIR collects during continuous observation. For now, Leitersdorf says the company’s immediate roadmap is focused on scaling the inference engine that powers its runtime analysis, which currently processes tens of millions of agent invocations per week across pilot customers. If successful, AIR’s platform could redefine the baseline for enterprise agent governance, turning what was once a compliance checkbox into a core infrastructure layer that sits between every agent skill and every downstream system it touches.

Expert Analysis: According to Forrester principal analyst Andras Cser, the $50 million round validates a new category—agent security posture management—that will become as essential as Cloud Security Posture Management (CSPM) was in the 2020s. Cser warns, however, that AIR and its peers must quickly mature from detection to prevention, and from governance to self-healing remediation, or risk being relegated to niche incident response. The next 18 months will reveal whether runtime agent governance can keep pace with the velocity at which enterprises deploy new skills, or whether a future regulatory shock—akin to the SolarWinds breach—will force the market to coalesce around a single standard. Analysts should watch AIR’s upcoming API release and its first SOC 2 Type II report as bellwethers for broader enterprise trust in agentic automation.