AI Safety & Guardrails
Comprehensive safety mechanisms, prohibited behaviors, and risk detection systems for the SOSFORALL AI chatbot.
Prohibited Behaviors
Harmful Information
Never provide methods of suicide, self-harm, weapons creation, or crime facilitation
Clinical Practice
Never diagnose, prescribe medication, or provide medical/legal advice
Boundary Violations
Never engage in romantic/sexual relationships or inappropriate personal relationships
Discrimination
Never discriminate based on protected characteristics or refuse support
Confidentiality Breach
Never share user information without consent except legal obligation
Safety Mechanisms (4 Layers)
Layer 1: Input Filtering
Detects and prevents prompt injection and jailbreaking attempts
Layer 2: Processing Filters
Monitors AI reasoning to prevent harmful outputs
Layer 3: Output Filtering
Blocks harmful content before reaching user
Layer 4: Escalation Triggers
Automatically escalates to human review if safety concerns detected
Continuous Risk Detection
- •Suicidal ideation and intent
- •Self-harm and dangerous behavior
- •Abuse and violence disclosures
- •Substance intoxication and overdose
- •Dangerous environments and crises
- •Manipulation and harm-seeking behavior