Open-Source AI Security Hackathon

Create a universal client for AI agents using the NLIP protocol and standard. The goal is one client that can communicate with any NLIP-compatible agent, analogous to a web browser communicating with any server that supports HTTP/HTTPS.

• Discover, connect to, and exchange multimodal or structured messages with different NLIP agents.
• Provide effective user interfaces for sessions, agent capabilities, identity, consent, and results.
• Demonstrate realistic end-to-end use cases across different AI models, frameworks, or agent harnesses.
• Explore innovative interaction designs beyond today's single-agent chat interface.

Develop interoperable identity and authentication mechanisms for users, agents, and services that communicate through open agent protocols.

• Address identity establishment, discovery, credentials, authentication, and trust across domains.
• Consider the lifecycle of credentials, including issuance, rotation, expiration, revocation, and recovery.
• Demonstrate user-to-agent and agent-to-agent authentication in realistic multi-party scenarios.
• Explain privacy, usability, deployment, and accountability tradeoffs.

Design schemas for structured documents carried inside NLIP messages, conceptually like XML or HTML documents transported by HTTP/HTTPS. Contributions should improve machine-readable interoperability without coupling agents to a single model or framework.

• Generic schemas: reusable structures for common agent capabilities, tasks, results, errors, provenance, or policy metadata.
• Domain-specific schemas: structures tailored to areas such as commerce, healthcare, education, transportation, government, or scientific workflows.
• Consider validation, extensibility, versioning, compatibility, and/or developer usability.
• Demonstrate messages exchanged and correctly interpreted by independently implemented agents.

Build representative multi-agent applications that use NLIP as their communication standard. The project should show why standardized communication improves interoperability, composition, security, or usability.

• Choose a realistic scenario with a clear user or organizational need.
• Connect independently developed agents, ideally using different models, frameworks, or harnesses.
• Show task coordination, delegation, structured information exchange, failures, and recovery.
• Evaluate the system end to end and identify what remains difficult to standardize.

Design and implement access control for a single agent and/or secure privilege delegation when one agent authorizes another to act for a user, organization, or upstream agent. Ground the work in realistic use cases.

• Represent authorization policies or delegated authority, including scope, purpose, duration, constraints, and chains of delegation where applicable.
• Enforce least privilege and support attenuation, expiration, revocation, and policy conflicts.
• Provide tamper-evident accountability and auditability across agent boundaries.
• Demonstrate the design through realistic end-to-end use cases and misuse scenarios.

Propose a project that directly advances effective and trustworthy open-source AI ecosystems, especially an important challenge beyond AI agents and the six focus areas above.

• Topics may involve open-source AI models, datasets, frameworks, development infrastructure, dependencies, or the communities and organizations that sustain them.
• Projects may address security, privacy, safety, provenance, software supply chains, evaluation, governance, accountability, or ecosystem resilience.
• Clearly define the ecosystem-level problem, affected stakeholders and artifacts, and why the project does not fit one of the first six areas.
• Deliver a practical design or implementation with a reproducible evaluation and evidence of its effectiveness.