Power, but no one's in charge


· 6 min read
Imagine appealing a decision made in milliseconds, but preserved nowhere in full. This is increasingly how automated systems govern access, rights, and allocation.
Much before it was written on paper, recited in courtrooms, and institutionalised, justice relied on keeping a record of what happened. Who promised what, who took what, who was present, who was harmed or awarded.
The oldest known written law is the Code of Ur-Nammu, inscribed on clay tablets around 2100 BCE in the area of present-day Iraq. The code is casuistic: if something happened (a crime), then there would be a consequence (a punishment), for example: “If a man has cut off another man's foot, he is to pay ten shekels.” To ensure that damage is recognised, recorded, and justice, by the fashion of the times, exacted.
What we consider justice has evolved over the centuries, but it is important to note that, back then, the fine was meant to cover lost earnings. Thus, on top of Ur-Nammu being a record of law, it is also an example of assigning monetary value to nature and to the functioning of the human body for earning a living.
Later, our justice systems were intrinsically built around recording, documenting, archiving, and accountability. All in an attempt to make truth outlive power. Even today, the rule of law depends on the durability of our records and whether those records can be trusted when decisions are contested, after the fact.
In our current information environment, going back to “the records” increasingly means inspecting automated decision-making, for example, in AI agent-to-agent interactions. It is at this juncture that our, however modern, assumptions about reliable records erode. A number of consequential decisions about human and environmental rights, access, risk, or allocation are no longer made directly by people or recorded in verifiable ways. They are subject to a kind of algorithmic evanescence.
Automated systems operate in real time, interact with one another beyond human control, and leave behind records that are partial, proprietary, or even unintelligible outside the systems and organisations that control them.
As a practical example, an automated credit decision can involve multiple models that evaluate an applicant’s creditworthiness. They can produce an outcome, such as a loan rejection or approval, in milliseconds. But when decisions need to be contested, our traditional legal tools become inadequate in determining how interacting algorithms arrived at their conclusions. It is then up to regulators and legal practitioners to ensure that citizens at large can understand and contest those decisions.
Automated systems practice delegated authority and will do so in ever greater volume. Models approve decisions and trigger actions that no single person can oversee in real time. They function like institutional actors, but faster. Even when human-encoded objectives and rules constrain the systems, they initiate consequences on their own.
When one automated decision becomes the input of another, the chain of responsibility spreads across different software, contracts, and organisations. This leaves no clear point at which accountability can be assigned.
The challenge becomes juridical. Legal systems are designed to assign responsibility to identifiable parties and are based on evidence that can be examined, challenged, and preserved without tampering. Automatic systems, without shared and verifiable records, put a huge strain on this “old-fashioned logic”. As we enter an era of fully autonomous multi-agent systems, this inadequate record-keeping will only scale exponentially. What emerges is a legal swamp.
What happens to local and international law when disputable records disappear? In recent years, this is what courts and regulators have struggled with.
In 2020, the UK Home Office's algorithmic visa-streaming system was withdrawn following a legal challenge and public scrutiny. In late 2025, the UK Home Office faced renewed formal complaints regarding its use of the IPIC and EMRT algorithms in immigration enforcement. Civil liberties groups argue that, just as in 2020, the decisions were made without transparent, contestable records, making it impossible for applicants to understand or challenge how the outcomes were produced.
In the European Union, the European Court of Justice has emphasised, in its interpretation of Article 22 of the General Data Protection Regulation, that consumers denied credit based on automated processing are entitled to clear and meaningful explanations of how those decisions were reached, even when algorithms involve proprietary data. Regulation, however, still assumes that explanations can be reconstructed after the fact. Increasingly, they cannot.
When applicants challenge these decisions, institutions may have internal logs showing “the system” acted within design parameters. The design parameters, however, may not ensure fair treatment or a reliable record. Reconstructing how multiple automated models interacted to produce a specific result remains difficult.
With the EU AI Act in force and being progressively phased in, high-risk systems become subject to transparency and contestability requirements by mid-2026. This is a step towards automated systems that produce continuous logs of decisions, enabling post-hoc scrutiny, risk monitoring, and oversight by official bodies. It does not, however, make them fully transparent or immutable.
As a general rule, systems designed around single points of control centralise power and memory. This, in turn, means that records are kept where they are most convenient for those in control, not where they are most accessible to those affected and those responsible for deciphering them. While mandatory logging standards and third-party verification help to a degree, they remain vulnerable to tampering or partial disclosure.
Distributed record-keeping challenges this architecture by design. It ensures that no single party can revise the past and that evidence of automated action exists outside the institution whose conduct might later be questioned. Blockchain technology has also, and probably rightly so, faced extensive criticism, from speculative excess to governance failures. If one were to look past these failed implementations, the underlying principle of tamper-resistant, shared records remains relevant in an age of diminished trust in public institutions.
In fact, researchers have begun experimenting with blockchain as part of multi-agent systems, recognising that accountability depends less on the accuracy of individual decisions than on whether actions can be reconstructed and verified after the fact.
At the same time, distributed ledgers can become heavy on the system, with high computational costs, complex governance, and inflexibility. As a result, less resource-intensive alternatives, like cryptographic logging, are gaining traction. Logging mechanisms act as “digital notaries”, producing continuous, tamper-evident records with hashes and signatures, and can be embedded directly into AI development pipelines. This, in turn, aligns the system with regulation from the get-go.
None of this, of course, promises moral clarity, let alone moral progress. Shared records do not prevent the harm from occurring, and automated systems can just as easily amplify injustice as prevent it. Furthermore, regulation always lags behind and often goes unobserved.
Without contestable records, we risk encoding power imbalances into infrastructure itself, beyond the reach of politics or law.
Decentralised record-keeping and cryptographic logs can be seen as a step towards more accurate institutional memory, something we can return to and verify. This becomes all the more important when automated decision-making directly affects rights and access, for example, in human-rights-related or environmental decisions.
Law, politics, and human discernment remain the foundation of justice. No technical architecture can substitute for them. What ledgers and logging offer is a verifiable record: a place for accountability when decisions are no longer made at human speed or scale.
As institutions become software, authorising automated decision-making without auditable records is a governance failure.
illuminem Voices is a democratic space presenting the thoughts and opinions of leading Sustainability & Energy writers, their opinions do not necessarily represent those of illuminem.
Track the real‑world impact behind the sustainability headlines. illuminem’s Data Hub™ offers transparent performance data and climate targets of companies driving the transition.
illuminem briefings

AI · Corporate Governance
Andrea Bonime-Blanc

Green Tech · AI
illuminem briefings

Power Grid · AI
The Wall Street Journal

AI · Ethical Governance
The Washington Post

AI · Public Governance
energynews

Green Tech · Sustainable Business