Hands Free: What LLM Driven Vulnerability Research Looks Like

Team82 reports discovery and disclosure of five critical vulnerabilities (three command injections, an out-of-bounds write, and an XSS: CVE-2025-64126 through CVE-2025-64130) in the Zenitel TCIV-3+ video intercom and demonstrates that an Anthropic Claude Opus 4.6 LLM agent, using tools like UPX and Ghidra via Claude Code, can reproduce and accelerate vulnerability analysis end-to-end in under ten minutes, highlighting the potential for LLM-based automated vulnerability research to lower the barrier to finding severe flaws in embedded devices.