Glassbox AI — Mechanistic Interpretability and AI-System Investigation
======================================================================

Copyright (C) 2026 Ajay Pravin Mahale <mahale.ajay01@gmail.com>

DUAL LICENSE NOTICE
-------------------

This repository is distributed under two licenses. The file list below is
the authoritative map; LICENSE-COMMERCIAL repeats the BSL file list.

  1. Business Source License 1.1 (see LICENSE-COMMERCIAL)
     Applies ONLY to these files:
       glassbox/compliance.py       — EU AI Act Annex IV report generation
       glassbox/circuit_diff.py     — CircuitDiff mechanistic diff (Article 72)
       glassbox/risk_register.py    — Persistent risk tracking (Article 9)
       glassbox/bias.py             — Bias/fairness analysis (Articles 10, 15)
       glassbox/audit_log.py        — Tamper-evident audit chain
       glassbox/cross_model.py      — Cross-model circuit comparison
       glassbox/v7/                 — AI-system investigation layer (trace,
                                      diff, group diff, interventions,
                                      investigation report, evidence packages)
       dashboard/v7_investigation.py — Investigation tab of the web demo

     cross_model.py, glassbox/v7/ and dashboard/v7_investigation.py are
     covered from 2026-10-11 and from the first PyPI release after 4.5.1.
     Copies obtained before that remain under the terms they were
     published with.

     BSL summary (not a substitute for the license text): free for
     non-production use and for production use inside your own organisation.
     Offering a commercial compliance, AI-governance, AI-investigation,
     debugging or observability service or product to third parties requires
     a separate commercial license. Each version converts to Apache-2.0 four
     years after its first public release.
     Contact: mahale.ajay01@gmail.com

  2. MIT License (see LICENSE)
     Applies to every other file in this repository, including the
     interpretability core (glassbox/core.py and the remaining glassbox/
     modules), tests, examples, experiments, scripts and documentation.

THIRD-PARTY COMPONENTS
-----------------------

This software builds upon the following open-source works:

  TransformerLens
    Author: Neel Nanda and contributors
    License: MIT
    URL: https://github.com/neelnanda-io/TransformerLens

  PyTorch
    Copyright (c) Facebook, Inc. (Meta Platforms, Inc.) and its affiliates.
    License: BSD-3-Clause (with additional terms)
    URL: https://pytorch.org

ACADEMIC ATTRIBUTION
---------------------

If you use Glassbox AI in academic research, please cite:

  Mahale, A. P. (2026). Causally Grounded Mechanistic Interpretability for
  LLMs with Faithful Natural-Language Explanations. arXiv:2603.09988.

  Wang, K., et al. (2022). Interpretability in the Wild: a Circuit for
  Indirect Object Identification in GPT-2 small. arXiv:2211.00593.

  Nanda, N., et al. (2023). Attribution Patching: Activation Patching At
  Industrial Scale. arXiv:2310.10348.

PATENT NOTICE
--------------
See PATENTS.md for the public-disclosure (prior art) notice relating to methods
implemented in this software.
