Show HN: Ave, a behavioral classification standard for agentic AI (github.com)

A new behavioral classification standard for agentic AI, called AVE, has been introduced to enhance security in AI/ML applications. Unlike traditional software vulnerability standards such as CVE and OSV, which focus on fixed products and versions, AVE addresses unique vulnerabilities associated with executable components like skill files and server manifests in agentic AI environments. It assigns stable identifiers to specific behavioral classes and scores them using the OWASP AIVSS framework, allowing security teams to map findings to familiar frameworks like OWASP MCP and MITRE ATLAS. The significance of AVE lies in its ability to tackle the emerging risks posed by agentic AI systems, which often operate without conventional safeguards like type checkers or sandboxes. By highlighting vulnerabilities such as prompt injection and dynamic tool use, AVE helps developers proactively scan their AI components for security flaws that could lead to critical exploits, data exfiltration, and other malicious activities. The accompanying bawbel-scanner tool serves as a reference implementation, facilitating easy integration into existing development pipelines to automate the vulnerability scanning process, thereby reinforcing security measures in AI deployment.