Extracting MITRE ATT&CK techniques from cyber threat intelligence (CTI) reports is an open-set, multi-label problem. Analysts need high recall so no techniques are missed, and high precision so unsupported techniques are not hallucinated. A new paper on arXiv proposes CTI2TTP, an evidence-grounded approach to this task.
The method is described as using "diverge-then-converge verification." While the abstract does not provide implementation details, the name suggests a process that first generates a broad set of candidate techniques and then verifies them to converge on a supported subset. The goal is to reduce both false negatives and false positives in automated threat intelligence analysis.
The significance of the work lies in directly addressing the trade-off between recall and precision. By framing extraction as a verification problem, CTI2TTP aims to produce more reliable technique annotations from CTI reports, which could improve downstream threat modeling and response.