Principles Library / Authority-matched accountability

Authority-matched accountability

Accountability Requires Authority to Influence Outcomes

Principle intent

Hold people accountable only for outcomes they have the authority, access, resources and control to influence, and keep accountability with a named human.

Read the full intent

That holds even when AI recommends or executes the work, because responsibility can't transfer to the system. Accountability works when responsibility matches the ability to shape the decisions behind the result. That needs clear boundaries on who may recommend, decide, approve, execute, override or reverse an action. Without them, authority drifts and responsibility gets deflected onto the system.

When it holds: what it does for you

Actionable ownership

Owners hold the authority and resources to change the outcomes they answer for.

Faster local decisions

Routine calls get made where the context is, without waiting on escalation.

Traceable AI decisions

Every AI-influenced outcome has a named person who can explain it and step in.

When it is ignored: from early signs to a lasting condition

Early warning signs · Visible now, cheap to fix

Teams answer for delays caused by dependencies they can't control.
Managers own delivery results but can't influence priorities, staffing or sequencing.
After failures, people argue about who owned the decision, since nobody said beforehand.

What happens over time · Accumulates, costly to reverse

Accountability is symbolic. People answer for outcomes that others, or agents, really decide.
Decisions slow down while people wait for permission or political cover.
Nobody can trace a failure to the decision behind it, so lessons don't stick.

The condition it becomes

Accountability Fragmentation · Primary

Accountability is assigned. Control is not.

How this principle feeds it

responsibility without control is how the library defines this condition.

Dependency Density · Secondary

The team owns the commitment. Nobody owns the gap.

How this principle feeds it

when a commitment depends on another team, the owning team answers for a gap nobody owns.

Governance Drag · Primary

Work is done. Approvals are not.

How this principle feeds it

work held up in an approval queue that adds delay without adding control is this condition.

Attribution Failure · Primary

We know what happened. Nobody can explain why.

How this principle feeds it

decisions made off the record can't be traced once outcomes arrive.

Oversight Erosion · Primary

The review process still runs. It stopped meaning anything.

How this principle feeds it

human review keeps running while the reviewers can no longer tell a bad agent call from a good one.

In agentic delivery

AI lets decision authority shift to agents while the accountability model stays the same. Agents recommend, approve, prioritize or execute consequential actions, and humans stay formally responsible for results.

The question is whether the accountable human has meaningful authority to understand, constrain, intervene in and change the decision process. Having a human in the loop doesn't settle it. Hierarchies distance people from decisions too, but agents do it faster and it feels more legitimate, because the agent made the call. Treating an agent's decision as inherently neutral removes the felt need for anyone to own it, and failures start getting framed as technical errors instead of governance failures. Unaccountable agents amplify whatever other condition is operating.

Where is your system on this path?

Diagnose

AI signs are on by default. Use the AI switch to exclude them.

1. Which of these early warning signs are you seeing?

Teams answer for delays caused by dependencies they can't control.
Managers own delivery results but can't influence priorities, staffing or sequencing.
Routine decisions keep getting escalated because authority is centralized or unclear.
People use informal channels to get around official decision structures.
After failures, people argue about who owned the decision, since nobody said beforehand.
Leaders answer missed targets with more reporting and sign-offs, leaving decision rights unchanged.
Nobody can say which decisions an agent may make, even as it gains capabilities. (AI)
People approve agent recommendations they can't explain or realistically challenge. (AI)
People answering for agent outcomes can't configure, inspect or override them in practice. (AI)
Ownership comes up only after incidents, and then it's "the model decided." (AI)

2. Over the last few months, have you noticed…

For each consequence: Not seen / Sometimes / Persistently / Not sure.

Accountability is symbolic. People answer for outcomes that others, or agents, really decide.
People cover themselves and push risks elsewhere instead of owning outcomes.
Decisions slow down while people wait for permission or political cover.
Shadow decision channels become normal, so nobody can see who really decided.
Nobody can trace a failure to the decision behind it, so lessons don't stick.
People defer to automation so often that their own judgment has weakened. (AI)

3. The condition it becomes

Based on your answers, recognize at most two conditions: No / Partly / Yes, that is us.

Accountability Fragmentation: Accountability is assigned. Control is not.

Dependency Density: The team owns the commitment. Nobody owns the gap.

Governance Drag: Work is done. Approvals are not.

Attribution Failure: We know what happened. Nobody can explain why.

Oversight Erosion: The review process still runs. It stopped meaning anything.

Take your Diagnosis to the next level by invoking AI Diagnostic Coaching tool.

What else have you observed? Optional. Shared only if you choose to start coaching.

Start a Diagnosis session: review and edit a draft containing your note, conditions you recognized in Step 3 (including whether they match partly), warning signs and consequence ratings before sending it. No scores or principle files are sent.

See what practices will help you

Act

Showing all 7 practices. Answer the Diagnose questions and the most relevant come first.

Match accountability to real authority

Give each outcome owner the authority and resources they need to actually influence that outcome.

Start by asking your team: Who is explicitly expected to own this outcome, and what decisions can they actually make?

Avoid · You’ll know it’s working when

Avoid

Treating accountability as pressure instead of system design
Assuming formal role ownership means practical authority exists

You’ll know it’s working when

Managers who own delivery results can shape the priorities, staffing and sequencing behind them.

Push routine decisions closer to the work

Move routine decisions to the lowest level that has enough context and control to make them.

Start by asking your team: What decisions are being escalated that could be owned closer to the work?

Avoid · You’ll know it’s working when

Avoid

Centralizing decisions to increase control while leaving accountability decentralized
Announcing more autonomy without changing actual decision rights

You’ll know it’s working when

Routine decisions get made close to the work, and people stop routing around official channels.

Make decision rights explicit

Spell out who may recommend, decide, approve, execute, override or escalate. Make ownership visible across human, automated and shared decision chains.

Start by asking your team: Who may recommend, decide, approve, execute, override or reverse automated actions?

Avoid · You’ll know it’s working when

Avoid

Confusing transparency with accountability
Believing explainability removes the need for ownership

You’ll know it’s working when

Decisions go through known channels, and ownership is clear before anything fails.

Name a human owner for every AI-influenced outcome

Assign a named person to every AI-influenced outcome. They must be able to understand, challenge and defend the decision, beyond just signing off.

Start by asking your team: Can the accountable person explain and defend this decision, not just point to who approved it?

Avoid · You’ll know it’s working when

Avoid

Assuming automation transfers responsibility
Treating AI recommendations as non-decisional because a human technically approved them
Blaming vendors, models or data instead of examining governance design

You’ll know it’s working when

Owners can explain the agent decisions they approve, and nobody hides behind "the model decided."

Build overrides people can actually use

Set up practical override, rollback and intervention for AI-driven actions. Tie escalation and learning paths to named roles instead of the system.

Start by asking your team: If this agent action goes wrong, who can stop or reverse it, and how fast?

Avoid · You’ll know it’s working when

Avoid

Believing post-hoc review provides meaningful decision authority
Treating AI failures as purely technical issues instead of governance failures

You’ll know it’s working when

People who answer for agent outcomes can inspect and override them in time.

Revisit authority boundaries as things change

Review accountability and authority boundaries regularly as structures, dependencies and AI autonomy shift.

Start by asking your team: As AI takes on more decision or execution authority, does human accountability still match actual control?

Avoid · You’ll know it’s working when

Avoid

You’ll know it’s working when

Everyone can say which decisions each agent may make, and that list keeps up with its capabilities.

Remove constraints that block real ownership

Find and remove structural constraints that leave people accountable for factors they can't meaningfully influence.

Start by asking your team: What resources, access or controls do they lack, and which dependencies keep them from influencing the result?

Avoid · You’ll know it’s working when

Avoid

You’ll know it’s working when

Teams answer only for delays and outcomes they can actually affect.