The next decade of governance technology can close the accountability gaps that institutional failure exploits, or widen them. The decisive variable is not the technology in isolation. It is the architecture of control, disclosure, review, and remedy within which the technology is deployed.
The Technology's Governance Potential
Artificial intelligence, shared ledgers, and real-time data systems change the distribution and cost of information. They can make records searchable, comparisons repeatable, and institutional activity easier to monitor. These capabilities may reduce the information asymmetries that allow failure to remain hidden and the administrative costs that make comprehensive oversight difficult.
But the same capabilities can strengthen the institution rather than its public. When an institution controls the data, the model, the access rules, and the appeal process, technology may enlarge its informational advantage. Surveillance becomes cheaper, classifications acquire an appearance of objectivity, and decisions can be reproduced at a scale that makes individual error a systemic condition.
Distributed ledgers illustrate the distinction. NIST describes blockchain systems as shared, tamper-evident and tamper-resistant ledgers. That can improve the traceability of recorded transactions. It cannot establish that the original information was true: the U.S. Government Accountability Office notes that a blockchain can preserve the integrity of information after entry without verifying that it was entered correctly. The ledger can secure the record and still secure an error. ([nist.gov](https://www.nist.gov/blockchain?utm_source=openai))
Accountability Is More Than Transparency
The usual argument for governance technology is transparency. But transparency is only the first component of accountability. A working accountability architecture requires at least four things: observability, so that conduct and outcomes can be examined; answerability, so that an institution must explain what it has done; contestability, so that affected people can challenge the data, reasoning, and authority behind a decision; and correction, so that an error can be reversed and its causes repaired.
Technology can improve observability without producing any of the other three. Publishing a database does not compel an official to answer for what it reveals. Disclosing an algorithm does not give an affected person the expertise, standing, or resources required to contest it. Recording an error immutably does not correct the error. Accountability depends on the institutions surrounding the information: auditors, courts, legislatures, inspectors general, journalists, civil society, and enforceable rights of review.
What the Record Shows
Ukraine's ProZorro system provides a concrete example of an observability gain. The official platform describes a fully electronic public-procurement system with open access to tenders, open-source code, an open application-programming interface, and a public analytics module. Those features allow people outside the procuring institution to inspect records and build independent tools around the data. They do not, by themselves, guarantee honest procurement or effective sanctions, but they enlarge the field of possible scrutiny. ([prozorro.gov.ua](https://prozorro.gov.ua/en/about?utm_source=openai))
The United States' federal spending-data regime demonstrates both the value and the limit of such visibility. The DATA Act requires agencies to publish spending information through USAspending.gov. GAO found improvements in the quality of reported data while also identifying continuing problems involving completeness, accuracy, data standards, disclosure of limitations, and data governance. Publication reduced one information barrier, but the reliability of the resulting accountability still depended on the quality and governance of what was published. ([gao.gov](https://www.gao.gov/products/gao-20-75?utm_source=openai))
The danger appears when institutional visibility expands without reciprocal rights for the people being classified. In 2020, The Hague District Court held that the Dutch legislation governing SyRI, a system used to identify risks of benefits, allowances, and tax fraud, violated Article 8 of the European Convention on Human Rights. The court found the system's operation insufficiently transparent and verifiable. The failure was not simply that government used data. It was that the legal and procedural safeguards did not justify or adequately constrain the resulting intrusion. ([rechtspraak.nl](https://www.rechtspraak.nl/organisatie-en-contact/organisatie/rechtbanken/rechtbank-den-haag/nieuws/2020/02/syri-legislation-in-breach-of-european-convention-on-human-rights))
Chicago's predictive-policing experience exposed a related mechanism. The city's inspector general found that police risk scores and tiers were unreliable, that personnel had not been properly trained, that access and use controls were inadequate, and that several versions of the models had not been evaluated. The report also found that the information had been made broadly available inside the department and shared outside it without sufficient guidance or monitoring. Technology increased the institution's capacity to classify people, while weak data governance and weak controls made the classifications difficult to trust. ([igchicago.org](https://igchicago.org/wp-content/uploads/2020/01/OIG-Advisory-Concerning-CPDs-Predictive-Risk-Models-.pdf))
The Architecture of Deployment
The governance question therefore begins before procurement and continues after deployment. Who defines the problem the system is supposed to solve? Who selects and tests the data? Which errors are measured, and whose injuries count as unacceptable? Who can inspect the model and its operating rules? Are affected people told that automation was used? Can a human reviewer disregard the output? Is there an independent body with the authority and technical capacity to audit the system? Can the system be suspended when its reliability or legality is in doubt?
These are not secondary ethical additions to an otherwise technical project. They determine the political function of the project. A model designed to assist a reviewable decision is different from a model whose output becomes the decision. An audit log controlled only by the institution being audited is different from one available to an independent overseer. A public dashboard is different from a right to obtain the underlying record, challenge its accuracy, and compel correction.
Formal regulation can establish part of this architecture. The European Union's AI Act, for example, specifies human-oversight requirements for high-risk AI systems, including measures enabling an assigned person to interpret an output and, where appropriate, disregard, override, or reverse it. But general rules cannot settle every operational choice about data access, procurement contracts, institutional incentives, staffing, or the design of appeals. ([eur-lex.europa.eu](https://eur-lex.europa.eu/eli/reg/2024/1689?utm_source=openai))
The recommendations of Australia's Royal Commission into the Robodebt Scheme offer a concise account of what operational accountability requires. For automated government decisions, the Commission called for a clear path to review, plain-language disclosure that automation is being used, access to business rules and algorithms for independent scrutiny, and a body empowered to monitor and audit automated systems. Those requirements connect visibility to contestability and correction rather than treating disclosure as an end in itself. ([robodebt.royalcommission.gov.au](https://robodebt.royalcommission.gov.au/publications/report))
Technology amplifies the governance arrangements around it. Visibility without answerability produces exposure, not accountability. Automation without contestability produces scalable power without scalable remedy. A system closes an accountability gap only when those it affects can see, question, and reverse its operation—and when an independent institution can enforce correction.
Discussion