Draft v0.1

Transparency

Transparency defines how clearly an AI system communicates its presence, purpose, behavior, reasoning scope, actions, and limitations within an experience. It helps users understand when AI is active, what it is doing, what it will do next, and what information or policies shape its output or actions. 2 1 3

Transparency is not about exposing every internal detail of a system. It is about providing the information users need to form an accurate mental model of the AI, verify outputs when needed, and make informed decisions about whether to rely on, guide, or intervene in its behavior. 2 1 3

As AI systems become more capable and more operational, transparency becomes more important—not less. Strong transparency mechanisms help calibrate trust, reduce uncertainty, prevent overreliance, and support safe adoption of more powerful assistive, agent-assisted, and autonomous experiences. 2 1 3

Principles

1. Make AI presence explicit

Users should be able to tell when they are interacting with AI, where AI-generated content appears, and when an agent or model is active in the experience. AI should not operate silently in ways that remove user awareness or make system behavior feel ambiguous. 2

2. Explain what the AI is doing

Transparency should help users understand the current state of the AI—whether it is retrieving information, generating a response, planning a workflow, or executing an action. Clear status and progress messaging helps users interpret behavior and anticipate outcomes. 2 1

3. Show what will happen next

When AI is recommending, planning, or acting, users should be able to understand the likely next step before they commit. This is especially important for agent-assisted and autonomous systems, where visibility into planned actions, triggers, and scope helps users maintain confidence and stay in control. 1 2

4. Communicate reasoning, evidence, and scope

Users should be able to understand the basis of an AI output at the right level of detail for the task. Depending on the experience, this may include source grounding, rationale, confidence cues, policy references, or a plain-language explanation of what information the AI considered and what it did not. 2 1

5. Be clear about limits and uncertainty

Transparency includes communicating what the AI can and cannot do, when output may be incomplete or uncertain, and when users should verify results before acting. This helps users develop an accurate mental model and prevents misplaced reliance on AI-generated outputs. 2 3

6. Preserve visibility after action

Transparency should not end once the AI has responded or acted. Users need persistent visibility into what happened, why it happened, and what data, tools, or policies informed the result so they can review outcomes, troubleshoot issues, and maintain accountability over time. 1 2 3

Guidelines

Clearly label AI-generated content and AI-driven actions

Use consistent visual and textual cues to show where AI is involved in the interface. Labels, indicators, and attribution help users distinguish AI-generated content from system content, user-authored content, or deterministic automation. 2

Make activation and autonomy visible

When users activate AI—or when an agent is activated on their behalf—make that state visible. For agentic experiences, communicate the system’s autonomy level, access permissions, triggers, and action rights so users understand the scope of what the AI can do. 2

Use meaningful status and progress messaging

Replace vague or decorative system messages with useful language that explains what the AI is doing. Messages such as retrieving sources, evaluating options, generating a draft, validating a configuration, or preparing an action plan are more helpful than generic “thinking” or anthropomorphic filler text. 2

Provide rationale in plain language

Where appropriate, explain why the AI made a recommendation, surfaced an insight, or selected a course of action. Explanations should be easy to find, easy to understand, and matched to the user’s context and expertise level. 2

Make sources and evidence available when verification matters

When users need to assess the reliability of an answer or recommendation, provide source links, references, policy citations, or other grounding mechanisms that help them verify accuracy and understand provenance. This is especially important in informational and decision-support experiences. 2 1

Communicate capability boundaries and limitations

Set expectations early and reinforce them during use. Users should understand the AI’s intended scope, the kinds of tasks it can support, where its reasoning stops, and what constraints or safeguards shape its behavior. 2 1 3

Show planned actions before execution when possible

For systems that perform multi-step or state-changing work, surface a preview or plan of intended actions before execution. This gives users a chance to inspect the sequence, understand impact, and decide whether to proceed. 1 2

Preserve an understandable history of AI actions

Provide logs, timelines, summaries, or activity histories that let users inspect what the AI did after the fact. This is especially important for operational and autonomous systems where visibility supports auditability, troubleshooting, and governance. 1 2 3

Make unavailable states and failures understandable

If an AI feature is unavailable, fails, or produces incomplete output, explain what happened and what the user can do next. Clear availability and error messaging prevents confusion and preserves confidence in the experience. 2

Transparency across AI capability models

Informational AI

Transparency focuses on clarity, source grounding, and scope. Users should know that AI is summarizing or answering, what information it drew from, and where they can verify or explore further. Interfaces should avoid implying certainty or authority beyond what the system can support. 1 2

Examples 1 - Documentation chat that cites relevant sources and shows where information came from. 1 2 - “Why is this happening?” explanations that make system context visible without overstating confidence. 1 2

Assistive AI

Transparency focuses on rationale, confidence, and choice. Users should understand why a recommendation was made, what alternatives are available, and what information influenced the suggestion before deciding whether to accept or modify it. 1 2

Examples 1 - Configuration recommendations with supporting rationale and clear accept, edit, or dismiss actions. 1 2 - Report generation assistance that shows what content was drafted, what data informed it, and what still requires user review. 1 2 - Step-by-step setup guidance that makes the purpose and consequence of each recommendation clear. 1

Agent-Assisted AI

Transparency focuses on legibility of plans, actions, and checkpoints. Users should be able to see what the AI intends to do, what tools or workflows it will invoke, where approvals are required, and what occurred at each step during execution. 1 2

Examples 1 - Guided remediation workflows that preview planned actions before execution. 1 - Multi-step configuration automation that shows progress, checkpoints, and status at each step. 1 2 - Ops orchestration flows with visible action history, interruptions, and rollback context. 1

Autonomous AI

Transparency focuses on policies, triggers, decisions, outcomes, and auditability. Users may not approve every action in real time, but they still need to understand the boundaries the AI operates within, what events triggered action, what changed, and how to inspect or intervene when necessary. 1 2 3

Examples 1 - Auto-scaling systems that show the thresholds, rules, and signals driving scale decisions. 1 - Self-healing systems that record what issue was detected, what remediation occurred, and what the resulting state is. 1 - Policy-driven traffic routing adjustments that expose governing rules, recent actions, and rollback or override paths. 1

What good transparency looks like

A transparent AI experience helps users answer five questions throughout the workflow: 2 1 3

Is AI active here? 2
What is the AI doing right now? 2 1
What will it do next if I continue? 1 2
Why did it produce this output or take this action? 2 1
What are its limits, and how can I verify or intervene if needed? 2 3

Summary

Transparency means making AI behavior legible enough for users to build an accurate mental model, make informed decisions, and maintain appropriate trust. As AI systems take on more responsibility, transparency must extend beyond labeling and explanation to include visible plans, clear scope, meaningful status, evidence, and post-action visibility. Designing for transparency helps users understand, verify, and safely work with more capable AI systems. 1 2 3