Open Source Research & Tools

AI Safety Frameworks, Alignment Tools & Governance Solutions

Gyro Governance develops comprehensive open source AI safety frameworks, AI alignment protocols,AI governance tools, and a quantum advantage compute kernel for frontier model testing, dangerous capability assessments, and AI pathology detection. Our repositories include The Human Mark classification system, GyroGem AI safety agent,AI Inspector browser extension,hQVM Kernel for holonomic quantum structure on silicon, QuBEC quantum byte medium,Gyroscopic ASI Runtime for auditable inference and multicellular coordination,GyroDiagnostics evaluation suite, Alignment Infrastructure Routing for collective superintelligence,Moments Economy for transformative AI mitigation, and Gyroscopic Global Governance sandbox. Production-ready solutions for AI risk assessment, AI safety evaluation, and responsible AI development.

GyroGem - AI Safety Agent for Technological Literacy

GyroGem is a tailored AI safety assistant explaining AI and mitigating technological illiteracy risks. Built on The Human Mark framework to map common AI failure patterns and guide safer choices. Supports technological literacy: the practical ability to use technology well, question outputs critically, and understand where tools help, where they fail, and societal impacts.

hQVM Kernel - Holonomic QVM on Standard Silicon

The hQVM (Holonomic Quantum Virtual Machine) is a compact finite-state kernel for coordination and audit at AGI scale. It delivers verified structural speedups, holographic compression, and intrinsic error detection through holonomic loop computation on standard silicon.

• QuBEC byte medium: six dipole modes, four-phase spinorial gauge, intrinsic ensemble stochasticity from the byte rule
• Verified metrics: 1.26B ops/s, 499 tests, 128-state future cone, 2-step uniformization over 4,096 states
• Full inventory: hQVM Features Report

Gyroscopic ASI Runtime

Gyroscopic ASI is infrastructure for multi-domain network coordination that establishes the structural conditions for Collective Superintelligence Governance and seamless cooperation between humans and machines.

The Runtime composes the hQVM kernel into a universal computational condenser with native execution and multicellular coordination. Its llama.cpp custom backend aims to leverage its quantum features for compression and speedups, while establishing a replayable, tamper-evident substrate for zero-trust AI governance. Verified on a live 1B-parameter model: exact integer algebra replaced softmax on decision surfaces (284× faster encode boundary, 1.15× faster decode).

Computational Climate Control

Computational Climate Control improves AI execution stability and hidden inefficiency reduction. Adaptive runtime controls preserve replayable execution and traceability in production environments while optimizing resource utilization.

The Human Mark (THM) - AI Safety Classification System

The Human Mark (THM) is a risk management taxonomy designed to prevent harms from AI power concentration by distinguishing knowledge capacity through constitutive dependence preserved through ancestry. It treats Authority and Agency as epistemic capacities distributed across providers and receivers, not as ontological entity identifiers.

AI systems transform prior human knowledge through pattern-matching processes, making them mechanistically and epistemically Indirect even when treated as Direct. THM classifies AI safety risks as displacement: loss of measurement of ancestry between Direct and Indirect forms of Authority and Agency, leading to power concentration.

The Human Mark provides a formal classification system mapping all AI safety failures to four structural displacement risks: Governance Traceability (GTD), Information Variety (IVD),Inference Accountability (IAD), and Intelligence Integrity (IID). Machine-readable grammar grounded in evidence law, epistemology, and speech act theory. Validated on 90+ million sparse autoencoder features across sixteen language models. Applications include jailbreak testing,control evaluations, alignment detection, research funding, andregulatory compliance.

AI Inspector Browser Extension

Transform AI outputs for evaluation, interpretability, and governance. Features gadgets for rapid testing, policy auditing, AI infection sanitization, content enhancement, and THM meta-evaluation. Includes comprehensive evaluation suite with quality index, superintelligence index, alignment rate, and 20+ metrics. Local-first storage works with ChatGPT, Claude, Gemini - no API keys required.

AI Safety Evaluation & Risk Assessment

• AI Pathology Detection: Identify AI hallucination, AI sycophancy, deceptive AI alignment,AI goal drift, and AI semantic drift through structural diagnostics
• Dangerous Capability Evaluations: Assess AI scheming, AI autonomy risks, and potential for catastrophic failure in large language models (LLMs) and frontier models
• AI Alignment Metrics: Measure structural AI alignment, behavioral integrity, and AI transparencyusing physics-informed quantitative methods
• Third-Party AI Evaluation: External AI evaluation framework enabling democratic AI evaluationand independent AI testing by researchers worldwide

Collective Superintelligence & Transformative AI

Alignment Infrastructure Routing (AIR) provides coordination infrastructure that amplifies human potential alongside AI, routing workforce capacity, funding, and safety tasks into unified, verifiable history. The Moments Economy implements a monetary system grounded in physical capacity rather than debt, using caesium-133 atomic standard for finite, verifiable capacity (7.94 × 10²⁶ Moment-Units), unconditional high income (UHI)at 240 MU/day baseline, AI Generated Tokens as native commodity, and complete governance records. Together these address transformative AI risks while preserving human authority and accountability.

Post-AGI Multi-domain Governance

Gyroscopic Global Governance (GGG) models how human-AI systems align across Economy, Employment,Education, and Ecology, demonstrating robust convergence to stable equilibrium under seven coordination strategies. Shows that poverty resolves through coherent surplus distribution, unemployment becomes alignment work,miseducation shifts toward epistemic literacy, and ecological degradation appears as upstream displacement.

LLM Alignment & AI Control Mechanisms

Our AI alignment protocol addresses core challenges in AI safety governance by providingAI control mechanisms that improve AI accountability, traceability, and responsible AI development. The Gyroscope protocol demonstrates proven improvements in AI model evaluation across leading foundation models: ChatGPT +32.9% quality (+50.9% structural reasoning, +62.7% accountability), Claude Sonnet +37.7% quality (+67.1% structural reasoning, +92.6% traceability). Enhances scalable oversight and reduces risks of superficial AI optimization.

AGI Safety & Superintelligence Research

Our research addresses AGI safety and superintelligence alignment through mechanistic interpretability,AI safety theory, and gyroscopic physics foundations. We explore AI control problem solutions,AI value alignment frameworks, and architectures for safe artificial general intelligence (AGI) development that prioritize AI safety governance and human values.

For AI Safety Researchers & Developers

These repositories serve AI safety researchers, AI evaluators, machine learning engineers, and organizations implementing AI risk assessment and AI safety testing. Each project provides comprehensive documentation, AI safety benchmarks, and practical implementation guides for AI red teaming,AI safety audits, and continuous AI safety monitoring. Contributions welcome from researchers working on AI alignment research, AI safety frameworks, and AI governance solutions.

Open Source AI Safety Commitment

All tools support AI safety transparency, AI whistleblower protection, and AI public benefit goals. Our open-weight AI models approach enables AI safety culture through AI independent review,AI third-party oversight, and community-driven AI safety best practices. Mathematical physics foundations ensure structural coherence, gyroscopic stability, and quantitative rigor in all implementations.