ConsultingWorkbench-backed AI security engagements — map, attack, defend, and prove your AI systems.
Scope a Review

SecEng Attack · Labs

Security Data Flow Canvas · Early access

Threat Models That Ship

Turn AI and product architecture into structured threat models, Jira-ready remediation, and Confluence-ready security design evidence.

Model trust boundaries, data flows, AI risks, controls, and evidence in one workflow built for product security teams.

Security Data Flow Canvas

AI System Threat Model

STRIDEDFD

User Browser

External Entity

LLM Gateway

AI Process

RAG Pipeline

AI Process

Vector Store

Data Store

Tool Executor

Process

Audit Log

Data Store

Trust boundary: AI inference layer

Identified risks

Prompt injection via user input

HIGH

Retrieval context leakage

MED

Excessive agent permissions

HIGH

Audit log tampering

LOW

4 risks · 2 controls mapped · Jira-ready

Live demo
6+AI risk templates
STRIDECompatible notation
Jira + ConfluenceAtlassian-native
Early accessAvailability

The problem

Threat models should not die in workshop notes.

Threat modeling is too often trapped in workshops, whiteboards, and stale diagrams. Engineering teams need clear Jira work. Security teams need review evidence. Leaders need a readable risk picture.

Diagrams go stale

Architecture diagrams do not keep pace with real system changes. Threat models built from them become unreliable within weeks.

Tickets lose context

Security action items divorced from the model lose the architectural context that makes remediation decisions tractable.

Evidence gaps at audit

Compliance and audit reviews need documented design rationale. Workshop notes and meeting decks do not fill that gap.

AI systems make it harder

Modern AI products add flows and failure modes that traditional architecture diagrams do not capture: prompt construction, retrieved context, model provider boundaries, tool calls, agent actions, memory reads, and side-effect outputs.

  • Prompt injection via user and retrieved content
  • Retrieval context leakage across tenants
  • Excessive agency from tool-enabled agents
  • Unsafe output handling and data exfiltration
  • Model provider trust boundary violations
  • Audit log gaps and side-effect traceability

Outcomes

From architecture to action.

01

Model the system

Sketch your architecture as a security data flow canvas: external entities, processes, data stores, trust boundaries, and flows.

02

Identify risks

STRIDE-compatible threat discovery across every flow and boundary. AI systems get first-class templates for prompt injection, retrieval leakage, and excessive agency.

03

Create Jira work

Convert risks into structured Jira issues with context, severity, ownership, and remediation guidance — not vague risk language.

04

Publish evidence

Export a Confluence-ready security design record with controls, evidence, risk summary, and reviewer sign-off.

Atlassian-native

Built for the tools your teams already use.

Threat models that live inside Jira and Confluence stay connected to engineering work. Risks become tickets. Decisions become design records. Evidence is always traceable to the model that produced it.

Confluence

  • Security design record page
  • Threat model snapshot and summary
  • Control and evidence tables
  • Architecture reviewer sign-off
  • Living document — updates with the model

Jira

  • Risk-to-ticket with full context
  • Severity, component, and owner fields
  • Linked back to the canvas model
  • Remediation guidance in description
  • Tracks to sprint and release

Who benefits

Security teams

Auditable evidence of design review, control coverage, and risk disposition.

Engineers

Clear Jira tickets with architectural context and remediation guidance — not abstract risk language.

Engineering managers

A prioritized security backlog that connects to sprints and release gates.

Leaders and auditors

An executive risk summary and a structured record of threat modeling decisions.

Compliance teams

Design-time evidence for SOC 2, ISO 27001, and ISO 42001 controls without a heavyweight GRC system.

AI systems

Threat modeling for RAG, agents, and AI-enabled products.

AI-specific risk surfaces

AI systems introduce new flows and failure modes that normal architecture diagrams do not capture. The canvas has first-class support for every layer of the modern AI stack.

Prompt construction layer

Prompt injection, template injection, context manipulation

Retrieval / RAG pipeline

Context leakage, cross-tenant retrieval, poisoning

LLM gateway and model providers

Trust boundary violations, model substitution

Tool calls and agent actions

Excessive agency, unsafe tool permissions

Memory and session state

State poisoning, session fixation, replay attacks

Output handling and side effects

Data exfiltration, unsafe code execution, SSRF

Audit and observability layer

Log tampering, traceability gaps, evidence loss

The canvas treats AI systems as first-class threat model subjects. Pre-built templates surface the risk taxonomy for LLM applications, RAG pipelines, and agent-enabled products — so teams do not start from a blank STRIDE spreadsheet.

AI template

LLM gateway review

Map prompt flows, system prompt exposure, user trust levels, and output routing. Identify prompt injection and model substitution risks.

AI template

RAG pipeline model

Canvas the retrieval path: query, embedding, vector store, context window, and response. Surface leakage and poisoning exposure.

AI template

Agent threat model

Model agent tool permissions, action chains, memory reads, and side effects. Identify excessive agency and unsafe tool use.

AI template

AI product launch review

Full pre-launch threat model with Jira remediation backlog and Confluence evidence record. Evidence of security review before ship.

Framework alignment

AI risk templates align with OWASP LLM Top 10, MITRE ATLAS AI adversarial techniques, and NIST AI RMF governance controls — so evidence produced connects to the frameworks your buyers and auditors already reference.

OWASP LLM Top 10
MITRE ATLAS
NIST AI RMF
ISO 42001

Features

Everything the canvas needs to produce evidence.

Security Data Flow Canvas

DFD-style notation with external entities, processes, data stores, trust zones, data flows, and risks on a single canvas.

STRIDE-compatible risk discovery

Structured threat identification across spoofing, tampering, repudiation, information disclosure, DoS, and elevation of privilege.

Trust boundary mapping

Visualize and reason about system perimeters, data classification boundaries, privilege zones, and external integration edges.

AI/RAG/agent risk templates

Pre-built threat templates for LLM gateways, RAG pipelines, tool calls, agent actions, memory, and model provider boundaries.

Jira issue generation

Create Jira issues directly from the canvas with component context, risk severity, ownership links, and remediation instructions.

Confluence evidence reports

Auto-generate structured security design records for Confluence: threats, controls, evidence, and approvals in readable format.

Controls and evidence tracking

Link mitigations and controls to specific risks and flows. Track evidence status as design decisions and testing results land.

Executive risk summaries

Produce a concise risk posture summary for leadership: threat count, severity breakdown, control coverage, and open items.

Workflow

Threat modeling in one flow.

1

Import or sketch architecture

Start from an existing Confluence diagram, an architecture description, or draw the system from scratch on the canvas.

2

Mark trust boundaries and data classes

Define perimeters, privilege zones, and data sensitivity across every flow and storage layer.

3

Generate or review threats

Use STRIDE templates and AI-assisted threat generation to identify risks, or walk through the model manually.

4

Map controls and evidence

Attach existing controls, record gaps, and link to test results or audit evidence directly on the canvas.

5

Create Jira issues

Convert open risks into Jira tickets with full context: component, flow, severity, owner, and recommended remediation.

6

Publish the Confluence design record

Export a structured security design document to Confluence. Living models update as the architecture evolves.

Differentiation

Not another diagram. Not another spreadsheet.

Most threat modeling tools produce artifacts that stop at the edge of the security team. This product creates a living model that drives engineering work, captures evidence, and connects to the systems teams already use.

vs. Diagramming tools

Diagrams do not drive remediation. There is no risk registry, no Jira integration, and no evidence output. They produce pictures, not work.

vs. GRC platforms

GRC systems are too heavy for design-time product security. They track compliance programs, not architecture-level threat models.

vs. AI chatbots

A chatbot generates threat ideas without durable context. There is no canvas, no trust boundary model, no Jira output, and no evidence record.

vs. Spreadsheets

Spreadsheets do not connect risks to architecture. They go stale immediately and produce no Jira work or Confluence evidence.

What makes it different

Structured security data-flow canvas — not a whiteboard, not a chat window.

Risks are linked to specific components, flows, and trust boundaries in the model.

Jira issues carry full canvas context: component, flow, severity, and remediation.

Confluence records are generated from the model, not written from scratch.

Evidence is always traceable back to the threat model that produced it.

Living models update as architecture evolves — not a one-time snapshot.

Use cases

Where security teams deploy it.

Use case

AI product launch review

Threat-model a new AI-enabled feature before launch. Identify prompt injection, data exposure, and model provider risks early.

Use case

RAG application threat model

Map retrieval pipelines, vector stores, embedding models, and output handlers. Surface retrieval leakage and context injection risks.

Use case

SaaS architecture review

Review multi-tenant SaaS systems for trust boundary violations, privilege escalation paths, and data flow risks.

Use case

Secure design review before release

Run a structured pre-release design review. Produce evidence that risks were identified and mitigated before ship.

Use case

SOC 2 / ISO 27001 / ISO 42001 evidence

Generate security design records and control evidence that support audit requirements without a heavyweight GRC system.

Use case

Product security backlog generation

Build a prioritized Jira backlog of security remediation work directly from the threat model — not from a spreadsheet.

Early access

Bring threat modeling into the engineering workflow.

What early access includes

Threat Models That Ship.

Join the early access list to be first to run a structured AI threat model for your system — with Jira-ready remediation work and a Confluence security design record.

Platform delivery

Web application

Hosted canvas available now for standalone use — no Atlassian account required.

Open canvas demo

Confluence macro

Embed the threat model canvas inside Confluence pages. Risks and evidence live next to your architecture docs.

Atlassian Marketplace

Jira integration

Create and track Jira issues directly from the canvas model. Risks become sprint-ready security work.

Request access

Advisory engagement

Run a structured threat model for your AI system.

Bring an architecture, a new AI product, or a pre-launch security review. We produce a structured threat model with Jira-ready remediation work and Confluence evidence — using the same canvas and workflow as the product.

New lab

Atlassian Threat Canvas maps threats, controls, decisions, and evidence inside Atlassian workflows