Tenable Inspector Screens AI Agents for Prompt Injection, Secrets, and Excessive Permissions
Tenable’s three-stage review process screens open-source AI agents for security flaws before they are listed on its CyberAgents Exchange directory.
Key points
- Three tools have passed the Exchange Inspector vetting so far.
- The review checks for prompt injection, exposed secrets and excessive permissions.
- Vetted agents are reported to cut threat hunting times by 75 per cent.
Tenable has introduced the Exchange Inspector to review open-source AI agents before they receive a vetted tag on its CyberAgents Exchange. The system evaluates community-built agents, skills and Model Context Protocol servers that are increasingly entering security operations centre workflows. According to Tenable, three tools have already passed this rigorous inspection process.
Root cause
The need for vetting arises because AI agents carry credentials and execute actions autonomously. Tenable notes that an agent’s blast radius is larger than that of a typical open-source library. The company draws parallels to software supply chain attacks, where registries trusting contributors by default led to security incidents. Openness makes the exchange useful but necessitates a review process to ensure safety before deployment in production environments.
Attack path
The Exchange Inspector tests for 15 types of security issues across three layers of the stack. These issues range from conventional flaws such as server-side request forgery and path traversal vulnerabilities to agent-specific risks. According to Tenable, the review specifically looks for excessive permissions and memory poisoning. The automated checks use Tenable One AI Exposure to screen for prompt injection and exposed secrets, while OpenAI GPT Cyber models assess the code and threat model.
Affected versions
The CyberAgents Exchange currently lists three Exchange Inspector-vetted tools. Two of these agents were built by Tenable, while the third was developed by Splunk. Contributors report that these specific agents accelerate threat hunting, remediation triage and cloud posture investigation. The directory covers a broad range of the security operations stack, including agents that monitor scanner health and assess cloud posture.
Mitigation
Tenable security researchers verify the runtime behaviour of each agent in a clean environment. This human verification step follows the automated check and the frontier model assessment. The process ensures that agents do not exhibit malicious or unsafe behaviour before they are made available to the wider community. This three-gate system aims to provide confidence that the agents will act as intended when deployed in operational settings.
What to do and how to stay safe: Tenable
- Review the source code of any AI agent or MCP server before deploying it in your environment.
- Monitor for excessive permissions or unexpected tool calls made by autonomous agents.
- Verify that agents do not expose secrets or credentials in their runtime logs.
- Wait for vendor-provided vetting tags or independent security assessments before integration.
Step-by-step guide: Stop Sensitive Data Leaks in AI Chatbots: Priority Controls
General security guidance from the Vector Bulletin newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What does the Exchange Inspector test for?
It tests for 15 types of security issues, including prompt injection, exposed secrets, SSRF, path traversal, excessive permissions and memory poisoning.
How many tools have passed the vetting?
Three tools have passed the Exchange Inspector vetting, including two built by Tenable and one by Splunk.
What is the benefit of using vetted agents?
Contributors report that vetted agents accelerate threat hunting, remediation triage and cloud posture investigation, with one tool cutting hunt times by 75 per cent.
