CRIBRUM

Background and requirement 

Sharing intelligence inherently carries risk. The MoD has a legal responsibility to ensure intelligence shared with other states does not lead to illegal or wrongful acts being committed. CRIBRUM was proposed to provide proof of concept that artificial intelligence (AI) can help Defence Intelligence (DI) manage this risk more effectively and efficiently.

To achieve this, TP Goup were tasked with developing a tool that could take training intelligence documents and analyse them, providing the user with information around the entities and legal risks the document contains.

The tool was designed to provide several outputs which could help Defence Analysts and professional DI legal advisors (LEGADs) make more efficient and accurate decisions when sharing documents, speeding up the intelligence sharing process and reducing legal risk.

Approach

• Create a tool written in Python that can read Microsoft Word documents.
• Identify a natural language processing (NLP) Python library called Spacy with a pretrained named entity recognition (NER) machine learning (ML) model.
• Develop a system of users entering their own entities and their associated legal risk.
• Design the tool to process intelligence documents, producing three outputs per input document: a data frame of entities and their category; a version of the input document with all entities highlighted according to risk level; and a document summary report giving metrics around the risk level and types of entity in the document.

Outcome

CRIBRUM successfully provided proof of concept that AI can be used to aid assessment of legal risk and help in the intelligence sharing process. The tool helped Defence Analysts/LEGADs minimise the legal risk of sharing intelligence documents with other authorities and make quicker decisions, streamlining the intelligence sharing process.