Tag definitions and scoring weight rationale for the AI Security Intelligence Engine.
Description: Security of autonomous AI agents, tool calling, function execution
Keywords: agent, agentic, tool calling, function calling, autonomous, multi-agent
Use Cases: Agent security frameworks, tool injection attacks, agent observability
Description: Prompt injection attacks, jailbreaks, adversarial prompts
Keywords: prompt injection, jailbreak, adversarial prompts, indirect injection, instruction injection
Use Cases: Prompt attack techniques, defenses, detection methods
Description: Human-in-the-loop designs, oversight, kill switches
Keywords: HITL, human-in-the-loop, human oversight, kill switch, circuit breaker, approval workflow
Use Cases: Human oversight patterns, approval workflows, safety controls
Description: Logging, tracing, monitoring, audit trails for AI systems
Keywords: observability, logging, tracing, monitoring, audit trail, tool call tracking
Use Cases: AI system monitoring, debugging, compliance auditing
Description: Unauthorized AI use, governance, asset inventory
Keywords: shadow AI, rogue AI, unauthorized model, governance, model registry
Use Cases: AI governance, asset management, policy enforcement
Description: Data exfiltration, privacy breaches, PII exposure
Keywords: data exfiltration, data leakage, information disclosure, PII exposure
Use Cases: Data protection, privacy controls, leakage prevention
Description: Model poisoning, supply chain attacks, checkpoint integrity
Keywords: model poisoning, supply chain, pre-trained model, checkpoint integrity
Use Cases: Supply chain security, model verification, dependency management
Description: Vulnerabilities, exploits, CVEs, security patches
Keywords: vulnerability, exploit, CVE, security patch, zero-day, critical flaw
Use Cases: Vulnerability tracking, patch management, exploit analysis
Description: Compliance, regulations, standards, governance policies
Keywords: regulatory, compliance, NIST, EU AI Act, OSFI, policy, governance
Use Cases: Regulatory compliance, policy implementation, standard adoption
What it measures: Topical match to priority AI security areas
Why this weight: Most important dimension - content must be relevant to be useful
What it measures: Source authority and research quality
Why this weight: Second most important - trustworthy sources produce reliable insights
What it measures: Severity, affected users, real-world consequences
Why this weight: Moderate importance - high-impact events warrant attention
What it measures: Recency of publication
Why this weight: Moderate importance - AI security evolves rapidly, recent content valued
What it measures: Actionable defender insights and mitigations
Why this weight: Important but not primary - not all valuable content has immediate mitigations
Edit config/scoring_weights.yaml to adjust dimension weights or tag keywords.
Weights must sum to 1.0 for proper normalization.