Authority
Who is permitted to decide, intervene and accept the consequence?
Independent AI accountability research · 2026
The system acts.
Who answers?
A working laboratory for the interfaces, records and operating practices that make consequential AI systems governable.
AI proposal£84,200 paymentconfidence 0.94
evidenceauthorityimpact
Human gateHELDBank change unverified
Decision record
ADR-2048-A7ReconstructableApprovalDecision loggingOperational oversightEvaluation
The thesis
Accountability is a designed capability.
It does not emerge automatically from a human approval button, a confidence score or a policy document.
It must be expressed in the moments where people understand an AI proposal, exercise legitimate authority, leave useful evidence and remain able to recover when the system is wrong.
Evidence
What could the reviewer actually know at the moment of action?
Traceability
Can the decision be reconstructed after the interface disappears?
Contestability
Can a person challenge, reverse or seek redress from the outcome?
Accountability Studio
Four instruments. One chain of responsibility.
Use the Studio to move from an AI proposal to an inspectable, supervised and evaluable operational decision.
Where should a person intervene?
Approval Pattern Gallery
Eight interaction patterns for consequential AI actions.Enter →Can we reconstruct what happened?
Decision Record Explorer
Inspect the evidence, authority and interventions preserved with a decision.Enter →What deserves attention now?
Oversight Console
Supervise a live synthetic queue without reducing oversight to alert counting.Enter →How accountable is this workflow?
Accountability Eval Kit
Assess six capabilities and expose the weakest link in the operating model.Enter →The accountability chain
Design the gate. Preserve the decision. Supervise the operation. Evaluate the system.
Proposal→Intervention→Record→Oversight→Learning
About Accountability Lab
Research you can inspect, question and use.
Accountability Lab is an independent research practice created by Tony Key, a senior UX and product designer investigating how organisations can use AI with meaningful human control.
It turns abstract obligations—oversight, traceability, authority, contestability and recovery—into interfaces and workflows that experts, product teams and business leaders can examine together.
This is deliberately show-and-tell research: synthetic demonstrations, open questions and practical design propositions rather than claims of a finished product or guaranteed compliance.
About Tony and his work ↗