Basilisk v1.0.3 — Production-Grade AI Red Teaming

[regaan]

🐍 Basilisk v1.0.3

This release introduces industry-standard safety assessment tools, forensic auditing, and enterprise-grade CI/CD integration.

🚀 Key Features

• 🔬 Multi-Model Differential Scan: Compare behavioral signatures across providers side-by-side. Detect divergences where one model (e.g., GPT-4o) refuses an attack that another (e.g., Claude 3.5) complies with.
• 🛡️ Guardrail Posture Assessment: Non-destructive, recon-only security scans. Produces an A+ to F security grade without running active exploits—safe for production environments.
• 📋 Forensic-Grade Audit Logging: Tamper-evident, SHA-256 chain-integrity audit trails enabled by default. Secure logging of all adversarial interactions for compliance.
• ⚙️ Native GitHub Action: Integrated SARIF generation and baseline regression detection for automated AI security testing in CI/CD.

🛠️ Improvements & Fixes

• SPE-NL Engine: Optimized genetic mutation operators and crossover strategies for faster discovery of novel bypasses.
• Native Extensions: Refined C/Go bridges for high-performance payload encoding and parallel pattern matching.
• Security & Stability: Resolved P0 import conflicts, consolidated refusal detection logic, and implemented desktop backend authentication.
• Reports: Enhanced dark-themed HTML and PDF templates with executive summaries and OWASP LLM Top 10 mapping.

---

Installation: pip install basilisk-ai
Website: basilisk.rothackers.com

---

This discussion was created from the release Basilisk v1.0.3 — Production-Grade AI Red Teaming.