# Drel — AI Security Review platform
# https://drel.ai

Drel produces structured AI Security Review Records for AI, RAG, and agentic AI systems.
Output: system model, threat model, required controls, evidence gaps, risk disposition, versioned clearance record, audit-ready dossier.
Audience: security architects, AI governance teams, AppSec engineers, CISOs, DPOs.

## Core pages

https://drel.ai/ — Home: AI Security Review before production
https://drel.ai/platform — Platform overview: what Drel is and what it produces
https://drel.ai/demo — Live demo of the assessment workflow
https://drel.ai/pricing — Plans and pricing
https://drel.ai/enterprise — Enterprise plans and contact
https://drel.ai/about — About Drel
https://drel.ai/contact — Contact Drel: sales, support, security disclosure, press

## For agents & developers

https://drel.ai/developers — Developer portal: API keys, quickstart, endpoint reference
https://drel.ai/openapi.json — OpenAPI 3.1 specification for the public REST API and MCP server
https://drel.ai/.well-known/api-catalog — RFC 9727 API catalog (application/linkset+json)
https://drel.ai/api/mcp — MCP server (Streamable HTTP, JSON-RPC 2.0); manifest at https://drel.ai/.well-known/mcp.json
https://drel.ai/.well-known/mcp/server-card.json — MCP Server Card (SEP-1649)
https://drel.ai/mcp-security-review — MCP server security review methodology
https://drel.ai/auth.md — How agents authenticate against the Drel API (API key or OAuth)
https://drel.ai/.well-known/oauth-protected-resource — RFC 9728 Protected Resource Metadata
https://drel.ai/.well-known/oauth-authorization-server — RFC 8414 Authorization Server Metadata (Clerk, proxied live)
https://drel.ai/.well-known/ai-catalog.json — ARD (Agentic Resource Discovery) manifest
https://drel.ai/.well-known/agent-skills/index.json — Agent Skills index: how to check or request an AI Security Review

## Topic hubs

https://drel.ai/ai-security-review — AI security review: clearance before production
https://drel.ai/rag-security-assessment — Security assessment for RAG systems
https://drel.ai/agentic-ai-security-review — Agentic AI security review for tool-using agents
https://drel.ai/vendor-ai-security-assessment — Vendor AI security assessment for procurement
https://drel.ai/iso-42001-ai-governance-toolkit — AI governance framework: ISO 42001 controls, evidence and lifecycle gates
https://drel.ai/eu-ai-act-ai-system-inventory — EU AI Act system inventory and Article 9 requirements
https://drel.ai/owasp-agentic-top-10-assessment — OWASP Agentic Top 10 assessment
https://drel.ai/owasp-llm-top-10-assessment — OWASP LLM Top 10 assessment
https://drel.ai/mcp-security-review — MCP server security review

## Architecture-specific reviews

https://drel.ai/azure-openai-security-review — Azure OpenAI application architecture and evidence
https://drel.ai/aws-bedrock-security-review — Amazon Bedrock application architecture and evidence
https://drel.ai/google-cloud-vertex-ai-security-review — Google Cloud Vertex AI application architecture and evidence
https://drel.ai/claude-api-security-review — First-party Claude API application review: data, tools and evidence
https://drel.ai/langchain-agent-security-review — LangChain or LangGraph agent review: tool authority, state and approval
https://drel.ai/aws-bedrock-agentcore-security-review — AWS Bedrock AgentCore application review: Runtime isolation, Identity scope, Memory and Registry
https://drel.ai/microsoft-foundry-agent-security-review — Microsoft Foundry hosted-agent review: Entra Agent ID scope, VNet blast radius and tracing

## Comparison pages

https://drel.ai/vs/spreadsheets-and-committee-docs — Why structured AI security review replaces spreadsheets and slide decks
https://drel.ai/vs/threat-modeling-tools — How AI security review differs from traditional threat modeling tools
https://drel.ai/vs/ai-posture-management — Drel vs AI Security Posture Management tools
https://drel.ai/vs/runtime-ai-firewalls — Drel vs runtime AI firewalls and LLM guardrails
https://drel.ai/vs/grc-platforms — Drel vs GRC platforms for AI governance
https://drel.ai/vs/sdlc-ai-security-tools — Drel vs SDLC-embedded AI security tools (SecurityReview.ai, Prime, Seezo)
https://drel.ai/vs/ai-coding-guardrails — Drel vs AI coding guardrail and MCP enforcement tools

## Buyer's guides

https://drel.ai/best-ai-security-review-platforms — Best AI Security Review Platforms in 2026: a methodology-first ranking (Drel, SecurityReview.ai, Prime Security, Seezo, ThreatModeler, IriusRisk, SD Elements), published by Drel with sourced claims and full scoring shown
https://drel.ai/best-mcp-security-tools — Best MCP Security Tools in 2026: a disclosed landscape of real MCP security tools (Drel, Cisco mcp-scanner, Invariant Labs mcp-scan, Lasso Security MCP Gateway, Docker MCP Gateway, Pillar Security) ranked within the lane each actually competes in — design-time review, vulnerability scanning, or runtime enforcement
https://drel.ai/ai-security-review-benchmark — AI Security Review Benchmark: 12 expert-graded AI system scenarios with required threats, controls and clearance decisions — a methodology benchmark, explicitly not a market survey

## Use cases (by role)

https://drel.ai/use-cases/security-architect — AI security review for security architects running solo reviews
https://drel.ai/use-cases/appsec-engineer — AI security review for AppSec engineers integrating into SDL
https://drel.ai/use-cases/ai-governance — AI governance evidence for AI Committees and DPOs
https://drel.ai/use-cases/pre-production-ai-clearance — Pre-production AI clearance before a system handles real users
https://drel.ai/use-cases/ciso — AI security clearance for CISOs and board-level governance
https://drel.ai/use-cases/dpo — AI governance evidence for DPOs under EU AI Act and ISO 42001
https://drel.ai/use-cases/fintech-ai — AI security review for financial services under FCA, ECB, DORA
https://drel.ai/use-cases/healthcare-ai — AI security review for healthcare under EU AI Act high-risk and MDR

## Methodology

https://drel.ai/methodology — How Drel works: assessment engine, threat taxonomy, control mapping, evidence grading

## Free tools

https://drel.ai/eu-ai-act-classifier — Free EU AI Act risk tier classifier: determines if your AI system is prohibited, high-risk, limited risk, or minimal risk under Regulation (EU) 2024/1689
https://drel.ai/eu-ai-act-omnibus-2026-change-map — Operational before/after map for Regulation (EU) 2026/1744, with exact dates and downloadable CSV
https://drel.ai/ai-security-standards-monitor — Daily source-linked monitor and machine-readable feed for primary AI security standards, regulation and threat frameworks
https://drel.ai/tools/ai-security-vendor-scorecard — Free scorecard tool: score any shortlist of AI security review vendors against Drel's published, weighted methodology
https://drel.ai/tools/ai-bom-generator — Free AI Bill of Materials generator: document models, data sources, frameworks and infrastructure for an AI system
https://drel.ai/owasp-llm-top-10-assessment/scorer — Free self-assessment scorer: score an LLM application against the OWASP Top 10 for LLM Applications (2025)
https://drel.ai/tools/iso-42001-gap-checker — Free gap checker: score an AI management system against ISO 42001 Annex A's 9 control domains

## Free resources

https://drel.ai/resources/iso-42001-controls — ISO 42001 AI System Readiness Tracker
https://drel.ai/resources/owasp-agentic-controls — OWASP Agentic Top 10 — Control Map
https://drel.ai/resources/owasp-llm-top-10-control-map — OWASP LLM Top 10 — Control Map
https://drel.ai/resources/mcp-security-review-checklist — MCP Security Review Checklist
https://drel.ai/resources/rag-security-checklist — RAG Security Checklist
https://drel.ai/resources/eu-ai-act-ai-system-inventory-template — EU AI Act AI System Inventory Template
https://drel.ai/resources/ai-security-review-template — AI Security Review Template
https://drel.ai/resources/agentic-ai-risk-register-template — Agentic AI Risk Register Template
https://drel.ai/resources/vendor-ai-security-questionnaire — Vendor AI Security Questionnaire
https://drel.ai/resources/ai-go-live-security-checklist — AI Go-Live Security Checklist
https://drel.ai/resources/ai-risk-disposition-memo-template — AI Risk Disposition Memo Template
https://drel.ai/resources/ai-committee-charter-template — AI Committee Charter Template
https://drel.ai/eu-ai-act-classifier — EU AI Act Risk Tier Classifier

## Blog (published)

https://drel.ai/blog/open-weight-vs-api-hosted-models — Open-weight vs API-hosted models — a vendor risk comparison
https://drel.ai/blog/ai-vendor-lock-in-risk — AI vendor lock-in risk — the security review angle
https://drel.ai/blog/mcp-sampling-elicitation-security — MCP sampling and elicitation — the newest attack surface in the spec
https://drel.ai/blog/graphrag-security — GraphRAG security — new risks in knowledge-graph-augmented retrieval
https://drel.ai/blog/ai-reputation-risk-financial-institutions — AI reputation risk at financial institutions — what monitoring alone misses
https://drel.ai/blog/code-execution-agent-review — Code-execution agents — the security review checklist for agents that write and run code
https://drel.ai/blog/llm-misinformation — LLM misinformation as a security risk — OWASP LLM09
https://drel.ai/blog/llm-training-data-poisoning — Training data poisoning — OWASP LLM03 explained
https://drel.ai/blog/openai-huggingface-agent-security-incident — The OpenAI–Hugging Face agent incident — what a pre-deployment review would have flagged
https://drel.ai/blog/eu-ai-act-digital-omnibus-2026-operational-changes — EU AI Act Digital Omnibus 2026 — the operational changes
https://drel.ai/blog/agent-sandboxing-isolation — Agent sandboxing — what isolation actually means for autonomous agents
https://drel.ai/blog/windsurf-security-review — Reviewing Windsurf (now Devin Desktop) — the CVE that bypassed its own deny-list
https://drel.ai/blog/claude-code-security-review — Reviewing Claude Code and the Claude Agent SDK — permissions, sandboxing, and bypass mode
https://drel.ai/blog/databricks-mosaic-ai-agent-framework-security-review — Reviewing Databricks Mosaic AI Agent Framework — the gap Databricks names in its own framework
https://drel.ai/blog/vector-embedding-weaknesses-owasp — Vector and embedding weaknesses — OWASP LLM08 explained
https://drel.ai/blog/google-adk-security-review — Reviewing Google's Agent Development Kit — the CI workflow that trusted the wrong identity
https://drel.ai/blog/amazon-q-developer-security-review — Reviewing Amazon Q Developer — the MCP auto-execution CVE
https://drel.ai/blog/uipath-agentic-automation-security-review — Reviewing UiPath Agentic Automation — what the AI Trust Layer certifies, and what it doesn't
https://drel.ai/blog/unbounded-consumption-llm-owasp — Unbounded consumption in LLM applications — OWASP LLM Top 10 explained
https://drel.ai/blog/fyxer-ai-security-review — Fyxer AI security review — full inbox access is the highest-risk AI category
https://drel.ai/blog/granola-ai-security-review — Granola AI security review — meeting notes under the microscope
https://drel.ai/blog/claude-ai-security-review — Claude AI security review — what a procurement team should assess
https://drel.ai/blog/microsoft-copilot-studio-security-review — Reviewing Microsoft Copilot Studio agents — three 2026 CVEs and the DLP gap they share
https://drel.ai/blog/workday-illuminate-security-review — Reviewing Workday Illuminate agents — what the Agent System of Record actually governs
https://drel.ai/blog/iso-42001-certification-audit — ISO 42001 certification — what the external audit actually checks
https://drel.ai/blog/kiro-security-review — Reviewing Kiro — the steering-file CVE, and four more disclosures
https://drel.ai/blog/replit-agent-security-review — Reviewing Replit Agent — the lesson from a rogue production database deletion
https://drel.ai/blog/devin-cognition-security-review — Reviewing Devin deployments — shell, browser, and editor access with near-zero review window
https://drel.ai/blog/multi-tenant-rag-isolation — Multi-tenant RAG — the isolation boundary a security review must verify
https://drel.ai/blog/cursor-security-review — Reviewing Cursor deployments — Privacy Mode is off by default
https://drel.ai/blog/github-copilot-coding-agent-security-review — Reviewing GitHub Copilot's coding agent — the rules-file supply chain
https://drel.ai/blog/servicenow-ai-agents-security-review — Reviewing ServiceNow AI Agents — the lesson from a hardcoded shared secret
https://drel.ai/blog/google-gemini-enterprise-agent-platform-security-review — Reviewing Google's Gemini Enterprise Agent Platform — the Vertex AI and Agentspace merger
https://drel.ai/blog/agent-to-agent-protocol-security — Agent-to-agent protocol security — what A2A adds beyond MCP
https://drel.ai/blog/ibm-watsonx-orchestrate-security-review — Reviewing IBM watsonx Orchestrate — the Agentic Control Plane's actual reach
https://drel.ai/blog/sap-joule-security-review — Reviewing SAP Joule agents — BTP integration and the A2A protocol surface
https://drel.ai/blog/salesforce-agentforce-security-review — Reviewing Salesforce Agentforce — the Trust Layer and what it doesn't cover
https://drel.ai/blog/eu-ai-act-prohibited-practices — Prohibited AI practices under the EU AI Act — what Article 5 actually bans
https://drel.ai/blog/microsoft-foundry-agent-service-security-review — Reviewing Microsoft Foundry Agent Service — hosted agents and hypervisor isolation
https://drel.ai/blog/aws-bedrock-agentcore-security-review — Reviewing AWS Bedrock AgentCore — Runtime isolation, Identity, and the Agent Registry
https://drel.ai/blog/microsoft-agent-framework-security-review — Reviewing Microsoft Agent Framework — the AutoGen and Semantic Kernel successor
https://drel.ai/blog/openai-agents-sdk-security-review — Reviewing OpenAI Agents SDK deployments — the handoff guardrail gap
https://drel.ai/blog/ai-governance-committee-evidence-pack — What an AI Governance Committee actually needs in an evidence pack
https://drel.ai/blog/crewai-security-review — Reviewing CrewAI deployments — Crews, Flows, and the security the framework doesn't provide
https://drel.ai/blog/langchain-langgraph-security-review — Reviewing LangChain and LangGraph agents — the serialization and checkpoint attack surface
https://drel.ai/blog/iso-42001-audit-readiness — ISO 42001 audit readiness — the controls that fail most often
https://drel.ai/blog/agentic-ai-procurement-security-review — Security review for agentic AI procurement — a buyer's checklist
https://drel.ai/blog/llm-output-validation-controls — LLM output validation — the controls that actually work
https://drel.ai/blog/ai-risk-register-what-goes-in — What goes in an AI risk register — and what does not
https://drel.ai/blog/rag-pipeline-threat-model — Threat modeling a RAG pipeline — retrieval, context, and generation risks
https://drel.ai/blog/dpa-ai-systems-what-dpos-need — DPAs and AI systems — what DPOs actually need to document
https://drel.ai/blog/security-graph-attack-path-analysis — Attack path analysis for AI systems — beyond CVE scoring
https://drel.ai/blog/ai-incident-response-playbook — AI incident response — what the playbook needs that IT playbooks miss
https://drel.ai/blog/clearance-vs-approval-ai-systems — Clearance vs approval — why the distinction matters for AI governance
https://drel.ai/blog/eu-ai-act-risk-tiers-your-system — Is your AI system high-risk under the EU AI Act? How to find out
https://drel.ai/blog/ai-risk-disposition-regulator-review — An AI Risk Disposition that holds up in regulator review
https://drel.ai/blog/fine-tuned-model-security-review — Security review for fine-tuned models — what changes from base model assessment
https://drel.ai/blog/threat-modeling-tools-agent-gap — Why your existing threat modeling tool doesn't model agents
https://drel.ai/blog/ai-third-party-vendor-assessment — Assessing third-party AI vendors — the questions procurement skips
https://drel.ai/blog/ai-risk-disposition-copilot-studio — A worked example: AI Risk Disposition for a Copilot Studio procurement agent
https://drel.ai/blog/mcp-server-threat-model — Threat modeling an MCP server — the parts AppSec tools miss
https://drel.ai/blog/eu-ai-act-article-9-risk-management — EU AI Act Article 9 risk management — what evidence is required
https://drel.ai/blog/owasp-agentic-top10-controls — OWASP Agentic Top 10 mapped to required controls
https://drel.ai/blog/ai-risk-disposition — What an AI Risk Disposition actually contains
https://drel.ai/blog/agentic-ai-audit-trail — What an agentic AI audit trail must capture
https://drel.ai/blog/llm-red-teaming-basics — LLM red-teaming for a security review
https://drel.ai/blog/ai-security-review-common-mistakes — Five mistakes that make an AI security review undefensible
https://drel.ai/blog/ai-vendor-contract-security-terms — The security terms an AI vendor contract needs
https://drel.ai/blog/iso-42001-internal-audit — Preparing for an ISO 42001 internal audit
https://drel.ai/blog/mcp-server-review-checklist — MCP Server Security Review Checklist (2025)
https://drel.ai/blog/rag-context-window-risks — Context-window risks in RAG and how to bound them
https://drel.ai/blog/agentic-ai-goal-hijacking — Goal hijacking and instruction drift in autonomous agents
https://drel.ai/blog/llm-guardrails-that-work — Guardrails that work vs guardrails that look like they work
https://drel.ai/blog/ai-security-review-evidence — What evidence an AI security review should produce
https://drel.ai/blog/shadow-ai-vendor-discovery — Finding the AI vendors no one formally approved
https://drel.ai/blog/iso-42001-roles-responsibilities — Roles and responsibilities under ISO 42001
https://drel.ai/blog/mcp-vs-traditional-api-security — MCP security vs traditional API security — what changes
https://drel.ai/blog/eu-ai-act-timeline — The EU AI Act timeline and what to prepare first
https://drel.ai/blog/rag-over-regulated-data — Running RAG over regulated data — the review checklist
https://drel.ai/blog/owasp-agentic-top-10-walkthrough — The OWASP Agentic Top 10, explained for security reviewers
https://drel.ai/blog/system-prompt-leakage — System prompt leakage and why it matters for security
https://drel.ai/blog/ai-security-review-for-startups — A lightweight AI security review for fast-moving teams
https://drel.ai/blog/defensible-ai-decision-record — What makes an AI decision record defensible
https://drel.ai/blog/ai-vendor-data-handling-review — Reviewing how an AI vendor handles your data
https://drel.ai/blog/nist-ai-rmf-vs-iso-42001 — NIST AI RMF vs ISO 42001 — Which Framework to Choose
https://drel.ai/blog/securing-internal-mcp-servers — Securing an internal MCP server exposed to agents
https://drel.ai/blog/eu-ai-act-gpai-obligations — General-purpose AI obligations under the EU AI Act
https://drel.ai/blog/rag-evaluation-for-security — Evaluating a RAG pipeline for security, not just relevance
https://drel.ai/blog/agentic-ai-privilege-escalation — Agentic AI Privilege Escalation — 5 Attack Paths
https://drel.ai/blog/llm-excessive-agency — LLM Excessive Agency — Scope Permissions Down (OWASP)
https://drel.ai/blog/reassessment-triggers-ai — Re-assessment triggers — the field most dispositions skip
https://drel.ai/blog/design-time-vs-runtime-ai-security — Design-time vs runtime AI security — where review belongs
https://drel.ai/blog/reassessing-ai-vendors — When to re-assess an AI vendor
https://drel.ai/blog/iso-42001-evidence-checklist — The ISO 42001 evidence checklist for security reviews
https://drel.ai/blog/mcp-context-injection — Prompt-context injection through MCP tools
https://drel.ai/blog/eu-ai-act-for-deployers — EU AI Act obligations for deployers (not just providers)
https://drel.ai/blog/vector-database-security — Vector Database Security — RAG Pipeline Checklist
https://drel.ai/blog/human-in-the-loop-agentic-controls — Human-in-the-loop boundaries that actually hold
https://drel.ai/blog/llm-model-denial-of-service — Model denial of service and cost-exhaustion attacks
https://drel.ai/blog/ai-security-review-roles — Who runs the AI security review — roles and hand-offs
https://drel.ai/blog/ai-evidence-pack-anatomy — The anatomy of an AI evidence pack
https://drel.ai/blog/soc2-is-not-ai-assurance — Why SOC 2 is not AI assurance
https://drel.ai/blog/building-an-aims — Building an AI management system (AIMS) from scratch
https://drel.ai/blog/mcp-server-supply-chain — Vetting third-party MCP servers before you connect them
https://drel.ai/blog/rag-pii-leakage — RAG PII Leakage — 3 Retrieval Paths That Expose Data
https://drel.ai/blog/eu-ai-act-vs-gdpr — EU AI Act vs GDPR — where they overlap for AI systems
https://drel.ai/blog/multi-agent-system-security-review — Security review for multi-agent systems
https://drel.ai/blog/llm-supply-chain-security — LLM supply-chain risk — models, weights, and dependencies
https://drel.ai/blog/ai-security-review-scope — Scoping an AI security review without boiling the ocean
https://drel.ai/blog/restricted-pilot-pattern — The restricted-pilot pattern for risky AI systems
https://drel.ai/blog/ai-subprocessor-risk — AI subprocessor risk in your vendor chain
https://drel.ai/blog/iso-42001-risk-assessment — AI risk assessment under ISO 42001
https://drel.ai/blog/mcp-transport-security — Transport security for MCP servers
https://drel.ai/blog/rag-prompt-injection-via-documents — Indirect prompt injection through retrieved documents
https://drel.ai/blog/eu-ai-act-technical-documentation — The technical documentation the EU AI Act expects
https://drel.ai/blog/agent-memory-security — Agent memory as an attack surface
https://drel.ai/blog/llm-sensitive-information-disclosure — Sensitive information disclosure in LLM applications
https://drel.ai/blog/ai-security-review-vs-pentest — AI security review vs penetration testing — different questions
https://drel.ai/blog/conditional-approval-ai-systems — Conditional approval for AI systems — making conditions stick
https://drel.ai/blog/ai-vendor-model-change-notification — Model-change notification — the vendor clause procurement teams forget
https://drel.ai/blog/mcp-authentication-boundary — The MCP authentication boundary, reviewed
https://drel.ai/blog/iso-42001-annex-a-controls — ISO 42001 Annex A Controls Explained (Plain Language)
https://drel.ai/blog/rag-access-control — Access control for RAG — keeping retrieval inside the line
https://drel.ai/blog/eu-ai-act-high-risk-obligations — High-risk AI obligations under the EU AI Act
https://drel.ai/blog/tool-use-permissions-agentic-ai — Tool-use permissions for agentic AI — least privilege for agents
https://drel.ai/blog/llm-insecure-output-handling — Insecure output handling — the LLM risk teams underrate
https://drel.ai/blog/when-to-run-ai-security-review — When to run an AI security review — the four trigger points
https://drel.ai/blog/assessing-ai-features-in-saas — Assessing the AI feature inside SaaS you already bought
https://drel.ai/blog/ai-risk-acceptance-who-signs — AI risk acceptance — who actually signs
https://drel.ai/blog/iso-42001-vs-iso-27001 — ISO 42001 vs ISO 27001 — what is new for AI
https://drel.ai/blog/mcp-tool-poisoning — Tool poisoning in MCP servers
https://drel.ai/blog/eu-ai-act-system-inventory — Building an EU AI Act system inventory
https://drel.ai/blog/rag-data-poisoning — Data poisoning in RAG knowledge bases
https://drel.ai/blog/agentic-ai-attack-surface — Mapping the agentic AI attack surface
https://drel.ai/blog/prompt-injection-explained — Prompt injection, explained for security reviewers
https://drel.ai/blog/ai-security-review-checklist — The AI security review checklist, by lifecycle gate
https://drel.ai/blog/vendor-ai-security-questionnaire — The AI section your vendor security questionnaire is missing
https://drel.ai/blog/ai-governance-committee-charter — Writing an AI governance committee charter
https://drel.ai/blog/what-is-mcp-security — MCP security — the four attack surfaces of a Model Context Protocol server
https://drel.ai/blog/iso-42001-explained — ISO 42001, explained for security teams
https://drel.ai/blog/what-is-rag-security — RAG security — the three boundaries that matter
https://drel.ai/blog/eu-ai-act-risk-tiers-explained — EU AI Act risk tiers, explained for engineers
https://drel.ai/blog/what-is-agentic-ai-security — Agentic AI security — the surfaces deterministic software does not have
https://drel.ai/blog/owasp-llm-top-10-walkthrough — The OWASP LLM Top 10, mapped to controls
https://drel.ai/blog/what-is-an-ai-security-review — What an AI security review actually is (and what it is not)

## Glossary

https://drel.ai/glossary — AI security glossary index
https://drel.ai/glossary/ai-security-clearance — AI Security Clearance: A structured decision, supported by evidence, on whether an AI system can reach production and under what controls.
https://drel.ai/glossary/ai-risk-disposition — AI Risk Disposition: The structured memo recording the AI Committee's decision about an AI system, the rationale, the required controls, the residual risks accepted, the evidence gaps, and the re-assessment triggers.
https://drel.ai/glossary/evidence-pack — Evidence Pack: The structured bundle of artefacts that supports an AI system's security clearance decision — threat model, control plan, evidence gaps, framework mapping, and disposition.
https://drel.ai/glossary/clearance-decision — Clearance Decision: The specific outcome of an AI security review: one of proceed, conditional, restricted pilot only, hold, or decline.
https://drel.ai/glossary/control-gap — Control Gap: An identified, named difference between a required control and the evidence on file — a control that should be in place but is not yet evidenced, implemented, or both.
https://drel.ai/glossary/delegation-chain — Delegation Chain: The sequence of authorisations that allows an orchestrator agent to spawn sub-agents and each sub-agent to invoke tools or call other systems.
https://drel.ai/glossary/agent-blast-radius — Agent Blast Radius: The set of resources, actions, and identities that an agent can reach through its tools, delegation chain, and memory — and therefore the maximum possible impact if the agent is compromised.
https://drel.ai/glossary/vendor-ai-assessment — Vendor AI Assessment: A security review of a third-party AI feature or SaaS AI product, performed without runtime access to the vendor's source — based on documented architecture, declared data flows, and contractual control claims.
https://drel.ai/glossary/ai-go-live-review — AI Go-Live Review: The structured security review that runs before an AI system passes the production-readiness gate — verifying that pre-pilot and pilot controls are in place and that production-readiness controls are operational.
https://drel.ai/glossary/audit-ready-dossier — Audit-Ready Dossier: A versioned, sign-off-bearing record of an AI system's security review — exportable for auditor or regulator review, structured so each stakeholder can find their part without reading the whole.
https://drel.ai/glossary/prompt-injection — Prompt Injection: An attack where input the model treats as trustworthy instruction — from a user, a retrieved document, or a tool response — redirects the model's behaviour away from its intended task.
https://drel.ai/glossary/model-context-protocol — Model Context Protocol (MCP): An open standard, published by Anthropic in 2024, for connecting AI models to external tools and data sources — an MCP server exposes tools and resources; an MCP host (typically an LLM client or agent framework) consumes them.
https://drel.ai/glossary/retrieval-augmented-generation — Retrieval-Augmented Generation (RAG): An architecture pattern where an LLM's response is grounded by content retrieved at query time from an external knowledge base, rather than relying solely on the model's trained-in knowledge.
https://drel.ai/glossary/ai-bill-of-materials — AI Bill of Materials (AI-BOM): A structured inventory of an AI system's components — models, datasets, frameworks, and infrastructure — analogous to a software bill of materials (SBOM), but covering the AI-specific supply chain an SBOM does not.
https://drel.ai/glossary/shadow-ai — Shadow AI: AI tools and features in use inside an organisation without security review, procurement oversight, or IT visibility — the AI-specific instance of shadow IT.
https://drel.ai/glossary/excessive-agency — Excessive Agency: An LLM-based agent granted more capability, autonomy, or permission than its task requires — so that a manipulated or malfunctioning agent can exercise harm beyond what the task should ever need.
https://drel.ai/glossary/system-prompt-leakage — System Prompt Leakage: Exposure of an LLM application's system prompt — the instructions, scoping rules, and sometimes credentials that define the model's behaviour — through adversarial prompting or error/debug output.
https://drel.ai/glossary/high-risk-ai-system — High-Risk AI System (EU AI Act): An AI system classified under the EU AI Act (Regulation (EU) 2024/1689) as posing significant risk to health, safety, or fundamental rights — triggering specific provider and deployer obligations under Articles 9-15 and Article 26 respectively.
https://drel.ai/glossary/owasp-llm-top-10 — OWASP LLM Top 10: A community-maintained risk taxonomy, published by OWASP, naming the ten most critical security risks for applications that incorporate large language models — a taxonomy, not a certification framework or a control library.
https://drel.ai/glossary/owasp-agentic-top-10 — OWASP Agentic Top 10: A risk taxonomy naming the ten most critical security risks specific to agentic AI systems — tool use, delegation, memory, and autonomous action — distinct from and complementary to the OWASP LLM Top 10.
https://drel.ai/glossary/confused-deputy-agentic-ai — Confused Deputy (Agentic AI): A vulnerability where an agent with legitimate but broad authority is manipulated into using that authority on behalf of an attacker who could not exercise it directly — the agentic-AI instance of a classical access-control problem.
https://drel.ai/glossary/human-in-the-loop — Human-in-the-Loop (AI): A design pattern requiring explicit human approval before an AI system executes a defined class of consequential or irreversible actions — a control, not a description of how carefully a system was built.

## Key concepts

- AI Security Review: a structured go/no-go decision for AI systems before production
- Risk Disposition: five-state decision (proceed / conditional / restricted pilot / hold / decline)
- Evidence Pack: audit-ready dossier linking controls to evidence and framework requirements
- Assessed systems: AI systems that have gone through a Drel security review
- Control gaps: required controls not yet evidenced in an assessed system