09. AI Safety / Security / Governance / Observability
NeMo Guardrails
Cisco (Robust Intelligence)
Azure Content Safety
watsonx.governance
Datadog
Dynatrace
OpenAI
Anthropic
Google Cloud
Hugging Face
Databricks
MicrosoftCross-cutting defenses across the AI lifecycle: direct/indirect prompt injection defense, guardrails, red teaming, regulatory governance, and distributed tracing.
9.1 AI Application Security & Adversarial Testing
Direct/indirect prompt injection defense, instruction hierarchy, jailbreak testing, NeMo Guardrails, tool privilege escalation prevention, and safe deserialization.
Prompt Injection Defense, Jailbreak Interception & NeMo Guardrails
Strict instruction hierarchy control (System > Tool > User), perplexity anomaly filters, delimiter tag hardening, NVIDIA NeMo Guardrails (Colang 2.0), and OWASP Top 10 for LLM v2.0 vulnerability mitigations.
NeMo GuardrailsCisco (Robust Intelligence)Tool Privilege Scoping, DLP Egress & Supply Chain Security
Dynamic least-privilege tool token scoping, streaming PII redaction (Presidio), sandbox egress network policies, and model weight malicious deserialization defenses (Safetensors / ModelScan).
Cisco (Robust Intelligence)9.2 Responsible AI, Safety & Model Risk
14 harm taxonomies, fairness and bias metrics, human-in-the-loop escalation, Model & System Cards transparency, and security incident response.
Harm Taxonomy, Fairness/Bias Auditing & Human Oversight
Multi-modal content moderation across 14 harm categories (hate/self-harm/harassment/CBRN), demographic parity and equalized odds fairness auditing, and human-in-the-loop escalation controls.
Azure Content SafetyOpenAIAnthropicGoogle CloudModel Cards, System Cards Transparency & Incident Response
Standardized Model & System Cards transparency reporting, intended use and out-of-scope boundary definitions, MITRE ATLAS threat mapping, and Coordinated Vulnerability Disclosure (CVD).
Hugging FaceOpenAIAnthropicGoogle Cloud9.3 AI Governance, Compliance & Audit Evidence
GDPR/CCPA privacy compliance, Zero Data Retention (ZDR), policy-as-code (Rego), NIST AI RMF 1.0, EU AI Act conformity, and automated audit evidence bundles.
Privacy Compliance, Zero Data Retention & Policy-as-Code
GDPR/CCPA compliance, Zero Data Retention (ZDR) enforcement, user consent ledgers, Open Policy Agent (OPA/Rego) declarative policy gates, and automated access control reviews.
DatabricksNIST AI RMF Mapping, Third-Party Model Vetting & Audit Evidence
NIST AI RMF 1.0, ISO/IEC 42001, and EU AI Act regulatory mappings, third-party model vendor vetting, tamper-proof automated audit evidence bundles, and exception management.
watsonx.governanceMicrosoft9.4 AI Observability, Quality & Incident Response
OpenTelemetry GenAI semantic conventions, end-to-end span distributed tracing, embedding drift (MMD/Wasserstein), online quality probes, and error budget burn rate alerts.
End-to-End Distributed Tracing & OpenTelemetry GenAI
Standardized OpenTelemetry GenAI semantic conventions, nested agent/tool/RAG span cascaded tracing, TTFT and TPOT latency breakdowns, real-time payload inspection, and interactive visualization.
DatadogDynatraceQuality Drift Monitoring, Embedding Drift & Incident Response
High-dimensional embedding drift detection via MMD & PSI, online quality and feedback probes, SLO error budget burn rate alerting, and cross-layer security incident response runbooks.
DatadogDynatrace