MIRABELLE uses LLMs to uncover hidden business process vulnerabilities
MIRABELLE reads ISO 9000 docs and process logs to flag logic flaws that cause delays
MIRABELLE, a new system from researchers Ben Falchuk, Himanshu Garg, Euthimios Panagos, and Sioan Zohar, targets a blind spot in enterprise risk: vulnerabilities buried in business process documentation. Just as code and hardware have flaws, business processes can harbor logic weaknesses—conflicting requirements, ambiguous specs, missing quality checks, or implementations that drift from design. These often surface as costly delays, quality failures, and compliance gaps. MIRABELLE automates discovery by processing ISO 9000/9001 documentation, user guides, work instructions, and process execution logs, then converting them into attributed graph representations that can be probed using graph and formal logic techniques.
Extracting that business logic from plain text is hard due to domain complexity and sheer volume, so the team turned to LLMs. Their experiments, detailed in arXiv:2608.04271, evaluate several LLMs across the full vulnerability-detection pipeline—from flagging grammatical and technical errors in short phrasings all the way to recovering complete process structure, including operation sequences, decision points, and input/output resources. The results highlight where LLMs accelerate MIRABELLE and where process complexity still demands human expertise. For enterprises, this points toward a future where process vulnerabilities are caught automatically, before they impact operations.
- MIRABELLE mines ISO 9000/9001 docs, user guides, and execution logs for business logic vulnerabilities
- Converts natural-language processes into attributed graph representations analyzed via graph and formal logic
- Evaluates multiple LLMs for tasks ranging from technical error-flagging to full process structure extraction
Why It Matters
Automating vulnerability discovery in process docs lets firms cut delays, costs, and quality failures before they hit operations.