v1.1.1: Licensing Update (Apache 2.0)

Licensing

This patch transitions the Anti-PLUTO framework from the Creative Commons CC BY-NC 4.0 license to the Apache License, Version 2.0.

This change was made to better align with standard open-source software practices, allowing for broader commercial use, modification, and distribution, while providing explicit patent protections and preserving attribution via the newly added NOTICE file.

Packaging

• Updated PyPI metadata classifiers to reflect the OSI Approved Apache Software License.
• Bumped framework version to 1.1.1.

v1.1.0: Provenance Tracking & Generation Seeds

This is a minor patch for antipluto.

• Standalone Provenance Tracking: Added random generation seeds directly into the data schema. This enables standalone provenance tracking and reproducibility of synthetic LLM cohorts by mapping generated emails back to their original generation state and seed
• Bumped framework version to 1.1.0.

v1.0.1: Documentation & Asset Rendering Patch

Patch Notes

This is a minor documentation and asset patch for PyPI compatibility.

• Documentation: Updated the Anti-PLUTO SVG logo to use an absolute raw URL source, ensuring it renders correctly on the PyPI package landing page.
• Packaging: Bumped project version to 1.0.1 for index indexing.

No core framework code or APIs were modified in this release.

v1.0.0: Initial Release - Anti-PLUTO Framework

Excited to announce the inaugural open-source release of Anti-PLUTO (Anti-Phishing Lexical Utilities and Threat Observation).

This release introduces a unified, extensible applied research framework designed to handle the entire lifecycle of modern email security research. From raw dataset ingestion and privacy-preserving masking, to synthetic threat generation, baseline benchmarking, and finally, threat classification.

Framework Capabilities

• Robust Data Ingestion & Preprocessing:
  • End-to-end pipeline to parse complex RFC 5322 MIME structures.
  • Automated HTML stripping, thread slicing, and Unicode normalization to ingest chaotic real-world corporate corpora.
• Privacy-Preserving PII Masking:
  • Two-stage lexical sanitization engine utilizing deterministic regex tokenization alongside high-accuracy spaCy NER (Transformer-backed).
  • Safely masks URLs, IPs, email addresses, names, and organizations to eliminate target leakage and anonymize datasets.
• Procedural LLM Cohort Generation:
  • Built-in interfaces for Azure OpenAI, Anthropic, and OpenRouter to procedurally generate massive, contextually fluent synthetic email cohorts (Benign & Phishing).
  • Traceability tools to track generated emails back to their human seed data.
• Validation, Telemetry & Benchmarking:
  • Integrated statistical validation tools to monitor dataset integrity.
  • Automated benchmarking suite to evaluate new datasets and models against legacy industry-standard engines (SpamAssassin & Rspamd).
• Reference Implementation: Dual-Branch Classification Engine:
  • Provides a state-of-the-art reference model for differentiating 4 email cohorts (Human Benign, Human Phishing, LLM Benign, LLM Phishing).
  • Utilizes a 100k-dim stylometric feature extractor fused with a 768-dim semantic branch (fine-tuned DeBERTa-v3).
• Production API Serving:
  • Asynchronous FastAPI backend capable of real-time inference (up to 7,497 msgs/s) with ONNX model export support for seamless deployment.

Installation

Anti-PLUTO is available on PyPI! You can install the core framework via:

pip install antipluto

For ML classification, generation, and API capabilities, install the optional dependencies:

pip install "antipluto[dev,classifier,api]"

Documentation

For comprehensive setup instructions, configuration details, and full CLI usage, please refer to the README.

Released under the CC BY-NC 4.0 License for academic and non-commercial research.