Public showcase of the GAP Ecosystem (GAPdrone & GAPbot). Highly autonomous swarm robotics powered by Edge AI (Hailo-8) and ROS 2. Built for EUDR compliance, Industry 5.0, and Dual-Use emergency response for a safer society.

A Responsible AI Stewardship License for AI Safety and Welfare

Cyber-focused network diagnostic suite — zero-telemetry speedtest, BGP path visibility, jitter persistence, hardened SOC-grade access (dual-use).

Misinformation detection and defense toolkit

中文网安防御问题误拒(over-refusal)评测基准：测安全模型是否误拒正当防御/安全教育问题，含真实危害对照组，可机器评分。补全 agent-safety-bench-zh 的另一半。

The Agent Meta-Language — a capability-safe language for agents that make agents. Private while dual-use safety, proof, and release boundaries mature.

Resilience, secure coordination, and safety simulation for autonomous systems

PDF text-layer steganography toolkit — hide messages in PDFs that look identical to humans but are extractable by software. For watermarking, AI ingestion-pipeline testing, CTFs, and research.

Convert raw cybersecurity documents into safety-filtered instruction-tuning data. Three-layer safety architecture (system prompt + regex validator + quality filter). Companion to the 4M-document secdata-raw corpus on Hugging Face.

SkyMarshal C2 — multi-agency drone operations dispatcher; 1st place in the SpaceShield Hack 2026 Dual-Use track.

Every story sells a wish — and every wish has a dark twin that wears its face (the cult, the conspiracy theory, the pickup artist's pitch). A browsable, citation-gated catalog of the wish-fulfillment engines stories run on, and the grifts that wear each one's face.

Quantum Systems — dual-use eVTOL fixed-wing reconnaissance drones (Vector, Trinity Pro, Reliant) and QBase 3D mission software

Biological dual-use content classifier using Constitutional Classifiers methodology — biosafety constitution, synthetic data pipeline, DeBERTa-v3-base classifier

A rule-governed, genome-writing-native biosecurity middleware that supervises agentic design AI. It inspects the plan a design agent produced, not just the final sequence, across five hazard axes and returns allow / flag-for-review / refuse with cited evidence and a signed design passport. Open-data, safe-proxy reference implementation.

Empirical dual-use risk assessment of protein language models (ESM-2) and structure-based design tools (ProteinMPNN)

Dual-Use Ambiguity Casebook: 30 structured cases at the AI-era biology dual-use decision boundary

Clinician-led LLM safety evaluation framework for mental-health contexts + reusable Claude agent patterns