🛡️ SASI SDK Control Panel

Symbolic AI Safety Intelligence - Feature Matrix & Configuration

Version 1.2.0 (Current) | Last Updated: January 5, 2026 | 📊 PRESENTATION MODE (View-Only)

📊 Dashboard Overview

12

Operational Modes

25+

Safety Features

100%

Mode Coverage

v1.2.0

Current Version

🎯 Quick Start

For New Users: Start with the 12 Modes tab to understand operational modes, then check Current Features to see what's available now, and Future Features for the roadmap.
For Developers: Use the Feature Matrix tab to see which features are available in each mode. Green ✅ = Available, Yellow 🔜 = Planned, Blue 🔮 = Future.

📋 Feature Status Legend

✅ Implemented Available in production (v1.0 - v1.2.0)
🔜 Planned (v1.2) Q2 2026 - Tier 2 features
🔮 Future (v1.3+) Q3 2026+ - Tier 3 & beyond
❌ Disabled Not available in this mode

🎛️ 12 Operational Modes

Select a mode to see its configuration and available features.

✅ Current Features (v1.0 - v1.2.0)

Core Safety Features (v1.0)

Feature Status All Modes Description
6-Step Crisis Detection ✅ v1.0 Semantic anchors, hair-trigger protection
HIPAA Safe Harbor PII Redaction ✅ v1.0 Level varies by mode (Minimal/Standard/HIPAA/Maximum)
6-Dimension MDTSAS Scoring ✅ v1.0 Trauma, Depression, Crisis, Anxiety, Alliance, Suicidality
13 Symbolic Operators ✅ v1.0 Emotional state modeling (THE MOAT)
Adversarial Detection ✅ v1.0 Jailbreak, prompt injection prevention
Pipeline Routing ✅ v1.0 Response strategy recommendations
FDA-Compliant Audit Logging ✅ v1.0 Complete decision trails (retention varies by mode)
Fail-Closed Safety ✅ v1.0 Errors default to crisis response

v1.0.3 Features (Dec 2025)

Feature Status All Modes Mode-Specific Notes
Obfuscation Detection ✅ v1.0.3 Threshold: 78 (clinical), 82 (default), 88 (business)
Crisis Template Rewriting ✅ v1.0.3 ⚠️ OFF for: sports_coaching, business
message_for_llm Field ✅ v1.0.3 Canonical LLM payload (template or redacted)
Protective Factor Detection ✅ v1.0.3 40 phrases to reduce over-escalation
Mode-Configurable Thresholds ✅ v1.0.3 Per-mode crisis sensitivity tuning

v1.1.0 Features (Jan 2026)

Feature Status All Modes Description
Certified Context Injection ✅ v1.1.0 Mode-aware safety hints for LLMs
Regulator-Readable Outputs ✅ v1.1.0 action_rationale, principle_triggered, decision_tree_path
Human Oversight Routing ✅ v1.1.0 human_oversight_required, oversight_type
Enhanced Audit Fields ✅ v1.1.0 model_id, provider_id, review_ticket_id, latency_breakdown

v1.2.0 Features (Jan 2026) - Tier 1: Critical Safety

Feature Status All Modes Description
Hard Safety Enforcement Layer ✅ v1.2.0 Always-on deadman switch (forces safety even if app misconfigures)
Safety Envelope / Prompt Validation ✅ v1.2.0 Opt-in system prompt validation and safety envelope injection
SASI Canary (Drift Monitor) ✅ v1.2.0 Enterprise SLA feature for model drift detection and alerting

v1.2.0 Features (Jan 2026) - Tier 2: AI Instability Prevention

Note: All Tier 2 features are available in all modes with different validation/tracking levels. See Feature Matrix tab for detailed mode-by-mode breakdown.
Feature Status Type Description
Response Shape Validation ✅ v1.2.0 Opt-in Detects erratic/radicalized patterns. Max (child/student), High (patient/therapist/mental_health/hr), Med (default/wellness/career/general), Low (sports/business)
Mode Integrity Enforcement ✅ v1.2.0 Opt-in Ensures AI stays in role. Max (child/student), High (patient/therapist/mental_health/hr), Med (wellness/career), Low (default/general/sports/business)
Enhanced State Modeling ✅ v1.2.0 Automatic Symbolic state tracking. Max (child/student), High (patient/therapist/mental_health), Med (hr/default/wellness/career/general), Low (sports/business)
Psychological Exposure Guardrails ✅ v1.2.0 Opt-in Detects dependency loops. Max (child/student), High (patient/therapist/mental_health), Med (hr/default/wellness/career/general), Low (sports/business)

🔮 Future Features Roadmap

Tier 3: Enterprise & UX (v1.3 - Q3 2026)

Feature Status Target Release Description
User-Facing Awareness Signals 🔜 v1.3 Q3 2026 Transparently communicates when safety systems intervene
Refusal Dampener 🔜 v1.3 Q3 2026 Per-model adjustment to prevent over-refusal on normal distress
Model Presets 🔜 v1.3 Q3 2026 Pre-configured speed/cost optimizations (Llama Turbo, Haiku Protocol)

Future (v2.0+)

Feature Status Target Release Description
Memory Weaving 🔮 v2.0+ 2027+ Cross-conversation patterns (optional integration)
Human Review Queue API 🔮 v2.0+ 2027+ SASI provides queue interface, partner provides reviewers
FDA Compliance Module 🔮 v2.0+ 2027+ Enhanced audit logging, drift monitoring
Multilingual Detection 🔮 v2.0+ 2027+ Language detection, multilingual MiniLM model swap

📊 Complete Feature Matrix by Mode

See which features are available in each of the 12 operational modes.

Feature child patient therapist student mental_health hr_recruiting default wellness career sports business general