Skip to main content
Loading navigation
Back to challenges

Mastra AI Data Breach Triage & Forensic Incident Agent

Financial institutions need automated incident response frameworks to analyze hacker breach logs and quantify regulatory notification obligations. Build a Mastra AI workflow with memory and document tool integration to analyze breach logs and calculate breach severity scores.

Machine LearningHosted by Vera
Sign in free to start

What you are building

Deploy Mastra AI workflow agents to scan exfiltrated record metadata and generate automated breach impact summaries.

How work is evaluated

Evaluation

Evaluates PII count detection accuracy and calculation of mandatory notification timelines.

Datasets

Public datasets, plus the private datasets you have access to.

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

deadline_check

Verifies correct reporting window calculation

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

record_count_mae

Mean absolute error on estimated stolen records count • target: 0 • range: 0-1000

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Build a Mastra AI agent workflow with step-by-step memory tracking

  • Analyze raw network intrusion logs to identify compromise of personally identifiable information (PII)

  • Automate reporting compliance based on state and federal cyber disclosure deadlines

Resources

Dataset notes

Network traffic logs and database exfiltration traces simulating the Apollo Global hacker attack.

Challenge setup

Evaluates PII count detection accuracy and calculation of mandatory notification timelines.

Challenge input

JSON log dump and enterprise jurisdiction list

Mastra AI

Provides structured workflow graphs and memory for TypeScript agents

Evaluated output

JSON with estimated_records_stolen, mandatory_notification_required (bool), deadline_hours

Evaluation checks
  • Verifies correct reporting window calculation
  • Mean absolute error on estimated stolen records count • target: 0 • range: 0-1000
Configured evaluation evidence
  • Benchmark: Cyber-Triage-Bench
  • Record Count Mae target: 0
  • 1 public reference case
Configured runtime
  • JavaScript execution harness
  • javascript sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

Action Space
  • Mastra AI · Required
  • Mistral Nemo · Optional
  • Hugging Face Inference API · Optional
Orchestration
  • Mastra AI · Required

Evaluation contract

  • deadline_check · Weight 1
  • record_count_mae · Weight 1

Metrics

  • record_count_mae · Target 0

Benchmarks

  • Cyber-Triage-Bench
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON log dump and enterprise jurisdiction list

    Output: JSON with estimated_records_stolen, mandatory_notification_required (bool), deadline_hours

    Execution harness: JavaScript

Evaluation tools

  • Mastra AI · Agent Framework

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Mastra AI Data Breach Triage & Forensic Incident Agent