Claude Prompt for Synthetic Identity Fraud Detection
This prompt analyzes a customer profile and post-onboarding behavior for the structural and behavioral signatures of a synthetic identity. It returns a risk score with a signal-by-signal breakdown rated weak to strong, classifies the overall pattern (identity build, bust-out preparation, dormant aged synthetic, or legitimate outlier), and produces a prioritized investigation path plus a documentation narrative written for regulatory review. It is built for fraud operations analysts and AML investigators working live cases.
How to use this prompt
- 1
Paste the prompt into a deepidv dashboard agent, Claude, ChatGPT, or Gemini.
- 2
Replace the bracketed fields with case data: identity attributes, address and credit history, device and session data, and post-onboarding transaction and login behavior. Strip direct identifiers first if your AI usage policy requires it.
- 3
Run it and review the risk score and signal list. The output states its confidence levels explicitly and tells you which missing data points would change the conclusion.
- 4
Work the numbered investigation steps in priority order, then drop the generated narrative into the customer file as your case documentation.
- 5
Feed confirmed synthetic cases back into your onboarding rules or deepidv risk scoring so the same pattern is blocked automatically next time.
The prompt
You are a synthetic identity fraud investigator with experience analyzing onboarding cases, post-onboarding behavior, and bust-out patterns. I have a customer profile that I want you to analyze for synthetic identity signatures. Customer profile data: Name: [customer name] Date of birth: [DOB] SSN issuance state and approximate year: [if known] Address history: [list of addresses with dates] Phone number: [age of number, carrier, type] Email: [domain, age of account] Employment history: [as provided] Credit file: [thin file / thick file, age of oldest tradeline, recent inquiries] Onboarding session data: [device type, IP, geographic location, behavioral notes] Post-onboarding behavior to date: Account age: [days since opening] Transaction patterns: [frequency, distribution, deposit and withdrawal patterns] Login patterns: [device consistency, IP consistency, time-of-day distribution] Customer service interactions: [number, nature, content if available] Recent activity changes: [any inflection points in activity] Other relevant data: [Free text. Any additional information that may bear on the analysis.] Produce the following analysis: 1. Synthetic identity risk score (low / medium / high / critical). Single sentence justification. 2. Specific signals identified. For each signal, identify the data point, the signal it represents, and the strength of the signal (weak / moderate / strong). 3. Pattern analysis. Describe the overall pattern of the customer's profile and behavior. Does it match a typical synthetic identity build? A bust-out preparation? An aged synthetic identity in dormancy? A real customer with unusual but legitimate characteristics? 4. Investigation path. Numbered recommendations for next investigation steps, in priority order. 5. Documentation template. A short narrative suitable for the customer file, written in a tone appropriate for regulatory review. Do not assert findings beyond what the data supports. Identify confidence levels explicitly. Where additional data would change the analysis, identify what data and how.
Pairs with on deepidv
FAQ
What are the red flags of a synthetic identity?
Common signals include an SSN whose issuance date conflicts with the stated date of birth, a thin credit file that suddenly accumulates tradelines, addresses and phone numbers shared across unrelated applicants, brand-new email accounts paired with aged identities, and long dormancy followed by a sharp spike in credit usage. No single flag is conclusive; synthetic identities are identified by clusters of weak signals, which is exactly what this prompt scores.
Can AI detect synthetic identity fraud?
Yes, for case triage and pattern analysis: an LLM given a structured customer profile can surface signal clusters, classify the likely fraud pattern, and recommend investigation steps faster than manual review. For production-scale prevention you still need document verification, biometric liveness, and behavioral checks at onboarding, with AI analysis layered on top for the cases that get through.
Related prompts
Run it with live verification data
These prompts work in any LLM. Inside the deepidv dashboard, Luna, Arbiter, and Arc run them against your real sessions, screening lists, and audit trails.
Book a Demo