What you will find here
Different works examining the same underlying accountability problem.My work is not concerned with whether AI appears intelligent, nor whether governance structures exist on paper and if organisations present themselves as responsible and well controlled. It is concerned with something much narrower and much harder.
When one AI-assisted decision is challenged months or years from now, what evidence will still exist to explain it?
That means not what the organisation intended, nor what its governance framework required, nor even what its dashboard showed at the time, but the evidence itself. The actual records that survive scrutiny. Can the chain of authority still be reconstructed. Is there proof that a human genuinely made the decision, on a defensible basis, using information that was available at the time and can that be produced afterwards.
Unlike most AI governance discussion, I treat accountability as an evidence problem rather than a governance performance exercise. Most organisations can describe how their systems are supposed to work. They can point to policies, controls, oversight structures and compliance processes. Far fewer can prove one exact outcome afterwards in a way that survives legal, regulatory, journalistic or public challenge.
My writing also treats confidence with caution. Fluent and authoritative AI outputs are not automatically reassuring. They are dangerous precisely because the system does not know when it is wrong and usually presents error with the same confidence as truth. The same pattern increasingly appears inside organisations themselves. Assured board updates, polished governance presentations and carefully managed compliance reporting can create the appearance of control long before genuine evidential accountability exists underneath.
A distinction runs throughout my work between governance appearance and evidential reality. Governance appearance is what organisations present: policies, committees, oversight structures, dashboards, reporting lines and compliance activity. Evidential reality is something much narrower and much harder: whether one exact outcome can still be reconstructed and justified afterwards from records, authority and retained evidence that survive scrutiny.
Policies are not proof. Committees are not reconstruction. Infrastructure is not justification. Compliance is not survivability. Most organisations have built the first, very few have built the second.
This test checks five personal markers: Participation, Information, Understanding, Judgement and Evidence to see if your individual decisions will hold up under hostile scrutiny.
It traces the complete chain of proof behind that specific outcome: Decision, Authority, Record, Evidence and Basis to see if responsibility can actually be fixed.
This 16-question test isolates your immediate exposure across five critical vectors: Product Status, Reconstruction, Traceability, Disclosure Readiness and Lifecycle Control.
The Director Accountability Test
Oversight habits built for a slower world fail when automated systems operate at immense speed and change continuously. If things go wrong, regulators and insurers look past the corporate structure to examine individual conduct.
DAREB© - What must be shown for a decision to stand.
Most accountability frameworks look at how a system is intended to work in general terms. DAREB does the opposite: it starts with a single real outcome affecting a single real person at one exact moment in time.
The EU PLD Exposure Test
The revised Product Liability Directive (Directive (EU) 2024/2853) applies strict liability directly to software and AI systems. If an outcome triggers a legal challenge, corporate compliance checklists will not protect you.