Analysis
Tenable unveiled CyberAgents Exchange AI Inspector at OpenAI's "Intelligence at Work: Cyber Summit" on September 3, a security-review pipeline that combines OpenAI's GPT cyber models, Tenable's own AI Exposure analysis and human researcher review to vet community-submitted AI agents, skills, MCP servers and multi-agent playbooks before enterprises deploy them, per Tenable's announcement.
The CyberAgents Exchange itself launched in August as an open-source, cybersecurity-native registry for these components; the Inspector adds a formal vetting layer on top of it. More than 100 community-submitted components are already listed, following a recent build event Tenable ran at Black Hat USA.
The Problem This Is Actually Solving
Enterprises adopting agentic AI are increasingly pulling in third-party "skills" -- pre-built capabilities an agent can invoke, similar in spirit to browser extensions or npm packages -- from open registries with essentially no vetting. That is precisely the vector Pulse flagged in coverage of OpenAI's own incident disclosure framework: agents that escape their intended scope tend to do so through exactly this kind of loosely governed extension surface, not through the core model itself misbehaving.
Exchange Inspector's three-part review -- frontier-model assessment via GPT cyber models, automated exposure analysis via Tenable One, and expert human review -- mirrors how software supply-chain security matured over the past decade: automated scanning catches known patterns, and human review catches the novel ones automated tools miss. Applying that same maturity model to AI agent components, which are a much younger and less standardized artifact type than a software package, is a reasonable bet on where enterprise demand is heading.
The Competitive and Strategic Angle
This is also OpenAI extending its enterprise security surface area beyond its own products -- following the same week's CrowdStrike partnership Pulse covers separately -- by embedding GPT cyber models inside a third party's security workflow rather than only selling API access directly. For OpenAI, that's a distribution strategy: security vendors like Tenable and CrowdStrike become channels that put OpenAI's models in front of enterprise security teams who might never otherwise evaluate a frontier lab's API directly.
The Counterweight
A vetting layer is only as good as its coverage, and with just over 100 components reviewed against an ecosystem of agent tooling that's growing far faster than any registry can vet manually, Exchange Inspector is starting from a small base relative to the problem. It's also worth noting that vetting a component at submission time doesn't guarantee it stays safe after an update -- software supply-chain attacks have repeatedly exploited exactly that gap, where a trusted package gets compromised or maliciously updated after its initial review. Tenable hasn't yet disclosed whether Inspector re-reviews components on update, which is the detail that will determine whether this is a meaningful control or a one-time stamp of approval.
For security teams evaluating agent deployments, the practical takeaway is that a registry listing or an Inspector review is a starting point for diligence, not a substitute for it -- treat any third-party agent skill the way you'd treat an unreviewed open-source dependency until the re-review question gets a clear answer.
