| Koinic Labs: Central Compliance & Transparency Report |
| Date: April 2026 |
|
|
| Status: SME Provider (Research & Development Phase) |
|
|
| 1. Copyright Policy (EU 2019/790) |
| Koinic Labs respects the rights of content creators. In accordance with Article 4(3) of Directive (EU) 2019/790, we honor all machine-readable reservations of rights (TDM opt-outs). Our training pipelines are designed to exclude data from sources that have explicitly opted out of AI training. |
|
|
| 2. Training Data Summary (Synthetic-First) |
| The AXL Architecture models are trained using a Synthetic-First Methodology. |
|
|
| Source: Data is primarily generated through high-fidelity AI-driven instruction sets and code-generation pipelines. |
|
|
| Categories: Programming logic (Python, C++, Rust, Go), multi-scale reasoning, and cybersecurity defense patterns. |
|
|
| Curation: Automated filters and human-in-the-loop (HITL) checks are used to ensure data quality and architectural alignment. |
|
|
| 3. Intended Use & Boundaries (Liability Protection) |
| To ensure safety and compliance, use of Koinic Labs models is subject to the following boundaries: |
|
|
| AXL-Secure & AXL-Debugger Series: |
| Intended Use: Defensive cybersecurity augmentation, code auditing, and vulnerability patching assistance. |
|
|
| Human-in-the-Loop: These models are designed to assist human experts. They are NOT intended for autonomous deployment in critical infrastructure (e.g., power grids, healthcare, transport) without human verification. |
|
|
| Forbidden Use: Any offensive cyber-operations or unauthorized intrusion testing. |
|
|
| 4. Environmental Impact |
| Koinic Labs prioritizes sustainability. By optimizing for CPU-first inference, our models significantly reduce the carbon footprint compared to standard GPU-intensive LLMs. |
|
|
| Training Efficiency: Typical runs average 0.0070 kg CO2. |
|
|
