Copilot is an AI-powered assistant integrated into the Darwinium fraud detection platform. It helps fraud operators analyze data, create queries, and identify patterns by leveraging machine learning insights and natural language processing to make complex fraud investigation tasks more accessible and efficient.

Integrate copilot into your daily fraud investigation workflow by asking it to analyze suspicious patterns, generate targeted queries for specific fraud types, co-author detection features, and provide insights on complex data relationships. Use it as a research partner to explore new attack vectors and validate your hypotheses with data-driven recommendations.

• Has count-level access to data on request
• Can review graph data based on any plots on the page
• can review event-level or the current frame of data you have open in your investigations view
• can read files in your workspace or make edits (these are permission-driven)

Copilot can recommend features from scratch or work with you as you are editing features. You can ask it to “add features that would help to identify account takeover” for example

Simply ask copilot to give you a query such as “give me a query that would help me to identify bot traffic”. It will automatically check what data it has available and will make suggestions for you. The query that copilot generates can then be run immediately or used in rules.

If you are already a Darwinium customer, copilot can be enabled free of charge for your organization. You may need to check with your organization’s AI governance policies to determine whether it should be enabled or if it is permitted.

• Copilot’s database layer is only capable of getting the count of events matching a particular query, however as a user you are able to expose data that you are seeing in portal. Our system is built upon OpenAI.
• We are implementing a data anonymization layer to minimize the exposure of personally identifiable customer data, however in the interim you should avoid asking copilot questions around specific people, addresses or personally identifiable information

• “Show me login attempts with unusual velocity patterns
• "Find transactions with mismatched device fingerprints"
• "Identify accounts with rapid password reset sequences"
• "Query for payment methods used across multiple accounts"
• "Find sessions with suspicious behavioral anomalies"
• "Show me failed authentication attempts from new locations"
• "Detect accounts with unusual transaction timing
• “help me author features that would help to identify credential stuffing”
• “analyze this behavioral identity graph on my page and give me insights on what you see”
• “describe the problems with the journey you currently see”