Episode Details
Back to EpisodesWhen AI Meets SQL: Inside CVE-2026-32622 and the Prompt Injection to RCE Pipeline
Published 3 weeks, 4 days ago
Description
In this episode, Mike and Alex break down CVE-2026-32622 — a critical vulnerability in the open-source SQLBot project that turned a helpful AI database assistant into a remote code execution pipeline through stored prompt injection.
Topics covered include:
- How the three-flaw attack chain works: from Excel upload to prompt poisoning to PostgreSQL command execution
- Why LLM-powered database tools create unique security risks that traditional defenses don't catch
- The broader pattern of prompt injection vulnerabilities in enterprise AI applications
- OWASP's defense-in-depth framework for securing LLM integrations
- Practical steps every organization should take: least privilege, input validation, output filtering, and structured prompts
- What this means for the future of AI-augmented developer and analyst tools