Secure Agentic System Design for Privacy-Centric Services
This challenge involves designing a secure, privacy-preserving multi-agent system for a hypothetical next-gen personal data management platform. The system needs to ensure user data is handled with utmost discretion, mimicking principles of zero-knowledge proofs where data is verified without being revealed. You will utilize AutoGen for orchestrating autonomous conversations among agents focused on system design and architectural choices. OpenAI Swarm will manage agent deployment and coordination in a decentralized fashion. The core will involve implementing A2A protocol for encrypted agent communication and MCP tool integration for controlled, auditable access to sensitive data stores, ensuring privacy by design and by default.
What you are building
The core problem, expected build, and operating context for this challenge.
Design a privacy-preserving multi-agent system using AutoGen, OpenAI Swarm, Claude Sonnet 4, and Mistral Large 2, with A2A protocol and MCP tool integration for secure personal data management.
How work is evaluated
Evaluation will assess the security and privacy aspects of the designed system, focusing on the correct implementation of A2A protocol, MCP integration, and the privacy-aware decision-making of the agents orchestrated by AutoGen and OpenAI Swarm.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Demonstrate a privacy-by-design approach in agent interactions and data handling.
Orchestrate agent teams using AutoGen and OpenAI Swarm for system design and verification.
Implement A2A protocol for encrypted inter-agent communication.
Develop MCP-enabled tools for controlled access to simulated privacy-sensitive data.
Design a secure, privacy-preserving multi-agent architecture.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
Evaluation will assess the security and privacy aspects of the designed system, focusing on the correct implementation of A2A protocol, MCP integration, and the privacy-aware decision-making of the agents orchestrated...
Challenge input
{'user_story': 'As a user, I want my personal data (health, financial) to be securely stored and accessed only when absolutely necessary, with my e...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{'architecture_diagram_text': 'Detailed textual description of components, data flows, security mechanisms, A2A/MCP points', 'privacy_compliance_re...
- Verify that A2A communications are simulated as encrypted and authorized, and that unauthorized requests ar...
- Ensure MCP-enabled tools correctly apply access controls based on simulated consent and privacy policies.
- Privacy Compliance Score target: 90
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- Verify that A2A communications are simulated as encrypted and authorized, and that unauthorized requests are denied.
- Ensure MCP-enabled tools correctly apply access controls based on simulated consent and privacy policies.
Metrics
- Privacy Compliance Score · Target 90
- Agent Collaboration Efficiency · Target 10
Workflow tasks
- Task 1
Input: {'user_story': 'As a user, I want my personal data (health, financial) to be securely stored and accessed only when absolutely necessary, with my explicit consent.'}
Output: {'architecture_diagram_text': 'Detailed textual description of components, data flows, security mechanisms, A2A/MCP points', 'privacy_compliance_report': 'Assessment of design against privacy principles'}
Execution harness: Python
- Task 2
Input: {'agent_request': {'agent_id': 'FinanceAnalysisAgent', 'data_type': 'spending_categories', 'user_consent': 'granted_limited'}}
Output: {'a2a_response': {'status': 'success', 'payload': {'category_1': 150, 'category_2': 200}}, 'mcp_audit_log': [{'tool': 'get_anonymized_data', 'status': 'success', 'access_level': 'limited'}]}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.