Active · 35
Agent identity and credential issuance (unique non-human identity per agent)
v1.2·0 use cases
Least-privilege scoping per use case
v1.1·0 use cases
Tool/API permission allowlisting
v1.3·0 use cases
Session-based credential expiry and rotation
v1.1·0 use cases
Role-based access tied to use case owner not just the underlying model
v1.0·0 use cases
PII/PHI detection and redaction in inputs and outputs
v2.0·0 use cases
Data residency and cross-border transfer restrictions
v1.2·0 use cases
Data retention limits for logs memory and embeddings
v1.1·0 use cases
Sensitive data classification (SOX HIPAA GDPR EU AI Act tiers)
v1.4·0 use cases
RAG source governance (what documents/data an agent can retrieve from)
v1.1·1 use cases
Data minimization enforcement (no unnecessary field collection/exposure)
v1.0·0 use cases
Tool call allowlist/denylist per use case
v1.2·0 use cases
Rate limiting on tool invocations
v1.0·0 use cases
Spend/budget caps on paid tool or API calls
v1.0·1 use cases
External system write-access restrictions (no direct prod DB writes etc.)
v1.1·1 use cases
Sandboxing for code execution actions
v1.1·1 use cases
File system and network access boundaries for agentic code execution
v1.0·0 use cases
Approval gates for financial transactions or irreversible actions
v1.2·0 use cases
Toxicity/harassment filtering on generated content
v1.3·0 use cases
Prohibited topic and jailbreak-resistance policies
v1.4·0 use cases
Brand/tone/voice compliance for customer-facing outputs
v1.0·0 use cases
Code security scanning on agent-generated code before execution
v1.1·1 use cases
Self-modification restrictions (agent editing its own prompts/config)
v1.0·1 use cases
Loop/runaway execution detection (ReAct loop timeouts max iterations)
v1.1·1 use cases
Emergency stop / kill-switch enforcement
v1.2·0 use cases
Prompt injection detection (direct and indirect e.g. from retrieved documents)
v1.5·1 use cases
System prompt leakage prevention
v1.2·0 use cases
Model/version pinning and change-control approval
v1.1·0 use cases
Adversarial input detection (jailbreaks obfuscated payloads)
v1.3·0 use cases
Full action/decision audit logging with traceability to use case and owner
v2.0·5 use cases
Periodic policy compliance review and re-certification per use case
v1.0·0 use cases
Incident response and rollback procedures for policy violations
v1.2·0 use cases
Explainability/decision-trail requirements for regulated use cases
v1.0·1 use cases
Third-party/vendor agent risk assessment (supply chain governance for external agents or plugins)
v1.0·0 use cases
Credentials/secrets detection with its own dedicated enforcement action distinct from generic DLP
v1.0·0 use cases
Draft · 17
Delegation boundaries (can an agent grant itself new permissions?)
v0.1·draft
Human-in-the-loop approval thresholds for sensitive actions
v0.1·draft
Cross-agent trust boundaries (agent-to-agent authentication)
v0.1·draft
Memory persistence controls (what an agent is allowed to remember across sessions)
v0.1·draft
Consent and purpose-limitation checks on data use
v0.1·draft
Action reversibility classification (read vs write vs destructive)
v0.1·draft
Cross-tool chaining restrictions (preventing tool A's output from silently triggering tool B)
v0.1·draft
Hallucination/factuality confidence thresholds
v0.1·draft
Bias and fairness checks on decision-influencing outputs
v0.1·draft
Copyright/IP filtering on generated content
v0.1·draft
Goal drift detection (is the agent still pursuing its assigned objective?)
v0.1·draft
Task scope boundaries (preventing scope creep mid-session)
v0.1·draft
Multi-agent escalation limits (sub-agent spawning caps)
v0.1·draft
Deception/manipulation detection in agent reasoning traces
v0.3·draft
Input sanitization for untrusted content sources (web email documents)
v0.1·draft
Real-time anomaly detection and alerting
v0.1·draft
Code-in-output detection as a distinct signal type separate from generic content/adversarial signals
v0.2·draft
