AI Accountability.

Assign responsibility for AI decisions

A harmful decision has no named owner.

Define developer responsibility

Builders are unclear which failures remain theirs after handoff.

Define deployer responsibility

The organisation running the system has no stated duty for its decisions.

Define user responsibility

Users are blamed or excused without a clear line for what they control.

Record AI decisions

Decisions are made and not stored in a form anyone can review.

Reconstruct agent actions

After the fact, nobody can replay what the agent did.

Preserve decision evidence

The inputs and reasons behind a decision are discarded.

Enable independent audits

An outside reviewer cannot inspect how decisions were made.

Challenge automated decisions

People affected by a decision have no way to dispute it.

Provide appeal mechanisms

A challenge has nowhere to go once it is raised.

Correct erroneous decisions

A wrong decision stays in force after it is found.

Escalate high-impact cases

Serious cases stay on the automated path.

Allocate liability for AI failures

Harm happens and no party is clearly liable.

Handle multi-party AI systems

Several organisations share the system and none owns the failure.

Address autonomous actions

Harm from an action the system took on its own has no liability path.

Establish remediation mechanisms

People who are harmed have no route to a fix or a remedy.

Track data origins

Training and input data cannot be traced to a source.

Track model versions

A decision cannot be tied to the model version that made it.

Track system actions

Actions are not linked to the system, time, and account that took them.

Reconstruct system history

The sequence of changes that led to the current system is gone.