How AIR’s $50M Raise Exposes the Unseen Risks of AI Infrastructure

The meteoric rise of artificial intelligence (AI) has brought unprecedented efficiencies to businesses, but it also comes with a hidden cost. As companies increasingly rely on AI agents to manage their systems, a new software supply chain is emerging – one that’s fraught with risks and vulnerabilities.

This issue is not unique to the tech industry; rather, it’s reminiscent of the early 2000s operating system vulnerabilities, where unsigned drivers could compromise an entire system. However, our current security protocols are far more stringent than those in place at the time. Despite this, AI infrastructure remains woefully unprotected.

The stakes are high because AI agents become increasingly autonomous and connected to enterprise systems, essentially opening a backdoor for attackers. Poisoning the content consumed by these agents can be just as damaging as directly attacking them. AIR’s solution is centered on monitoring and vetting the skills and add-ons used by AI agents, providing visibility into agent activity, identifying rogue employees or unapproved tools, and blocking malicious actions in real-time.

AIR’s platform isn’t alone in this space; competitors like Noma Security, Zenity, Astrix Security, and Operant AI offer similar solutions, backed by significant venture funding. This suggests that the industry is finally acknowledging the gravity of these risks. However, it also raises questions about the long-term sustainability of this approach.

AIR’s CEO believes its moat lies in continuous vetting of the skills and add-ons ecosystem. Yet, can a single company truly keep pace with the rapid evolution of AI tools? The answer is likely no – at least not without significant investment in research and development. And yet, Saban remains optimistic that companies will still want to purchase an independent product that works across vendors.

This brings us back to the larger issue: are we treating AI as a security problem or an infrastructure one? The former suggests a focus on scanning and detection tools, while the latter implies a more fundamental shift in how we approach AI’s role within our systems. Bogomil Balkansky’s statement – that inspecting every skill and plugin is an “infrastructure problem long before it is a security problem” – highlights this tension.

AIR’s funding is a vote of confidence in its solution, but it also underscores the industry’s uncertainty about how to address these risks. As we continue down this path, one thing is clear: AI’s invisible risks will only intensify unless we develop more robust and adaptive solutions. The question is, will companies be willing to invest in infrastructure that keeps pace with the rapidly evolving AI landscape?

AIR has around 40 employees – a relatively small team for such an ambitious endeavor. Its ability to scale its solution, not just technically but also culturally, will be crucial in determining its success. The $50 million raised is a start, but it’s only the beginning of a much larger conversation about what AI security looks like in practice.

As we move forward, one thing is certain: the invisible risks lurking within our AI systems won’t disappear on their own. It’s time for companies to take responsibility for these challenges and invest in infrastructure that keeps pace with the rapidly evolving AI landscape. Anything less would be a replay of the early 2000s operating system vulnerabilities, but this time with far greater consequences.