In 2026, Artificial Intelligence (AI) is no longer a distant promise; it is a ubiquitous tool in public management. Technically defined as the capacity of computer systems to process data, recognize complex patterns, and generate autonomous responses (per OECD and DataCamp frameworks), its value proposition is clear: a productivity revolution designed to unblock historical administrative bottlenecks.

Far from being a technology to reject out of fear, AI has established itself as an everyday reality to which public institutions must adapt. While its utility spans many fields, it is particularly strategic for public administration—a sector traditionally strained by heavy administrative burdens.

This transformation is anchored in functions that are already operational, such as automated document processing, technical report synthesis, and advanced decision support (EsadeEcPol). However, it is precisely within decision support that the most profound ethical and operational dilemmas arise.

When we move from routine case processing to evaluating bids for public tenders or grant applications, the stakes shift dramatically.

The Debate: Who Evaluates Whom?

The core discussion is not whether the technology works, but whether it can navigate technical discretion without triggering a public crisis of confidence.

We are witnessing an unprecedented paradox: private companies use generative AI to draft pristine, perfectly compliant proposals, while government agencies delegate the task of grading those machine-written bids to algorithmic systems.

This dynamic raises critical questions about public sector integrity:

• The AI Feedback Loop: Are we evaluating the real-world quality and viability of a public service, or simply measuring which company uses better drafting algorithms than the evaluating agency itself?
• The Erosion of Trust: How can citizens trust an algorithm to award a multi-million-euro contract or a vital subsidy if the reasoning behind that decision is locked inside an opaque "black box"?

As the OECD warns, efficiency gains cannot serve as a blank check. Without strict regulation and clear governance, using AI in high-stakes procedures risks causing transparency failures that the public will not tolerate. The fundamental question is whether AI should act as a judge or merely an assistant, and how to guarantee that a human remains capable of explaining why one proposal won over another.

Proposal Evaluation as High-Risk AI

The debate over AI in proposal evaluation is not merely a technical preference; it is a legal imperative that impacts institutional legal certainty. Under the European Union Artificial Intelligence Act (AI Act), using AI systems to assess eligibility for public benefits, grants, or procurement bids is explicitly classified as "high risk."

This classification is deliberate. These systems operate at a critical intersection where an algorithmic score can alter access to public funding and fundamental rights.

The OECD stresses that automating these workflows without deep safeguards introduces severe institutional risks:

• Erosion of Accountability: If a systemic error or unfair decision occurs, responsibility becomes diluted among software vendors, data managers, and evaluating officials, complicating public accountability.
• Algorithmic Bias and Discrimination: AI models can perpetuate historical dataset biases. In a procurement or grant process, this can result in the unfair exclusion of specific companies or demographics based on patterns mistaken by the AI as "high risk."
• The "Black Box" Problem: The complexity of modern models makes it difficult for officials to explain why a proposal was rejected. If an agency cannot provide a clear, understandable justification, it violates the applicant's due process rights.
• Automation Bias: Evaluators risk becoming overly reliant on AI outputs, blindly trusting algorithmic scores and reducing human oversight to a rubber-stamp formality.

Control as the Foundation for Implementation

Crucially, the EU AI Act's "high-risk" classification is not a ban. Instead, it is an framework for the state to reinforce its role as the ultimate guarantor of public rights. The goal is not to avoid AI, but to define how much authority is delegated and ensure human oversight remains decisive.

Governance models must make technology fully auditable and transparent. Oversight should scale in direct proportion to the level of autonomy granted to the algorithm:

• Evaluative AI Approach (Maximum Risk): Exemplified by pilot projects like the UK tax authority (HMRC), which used the Outmatch system to screen and grade candidate responses in a fully automated manner. As the machine assumes decision-making weight, standards for transparency, security, and auditability must be absolute to prevent hidden bias.
• Assistant AI Approach (Lower Risk): Exemplified by the North Carolina (USA) Department of Information Technology, which launched a chatbot to provide technical assistance to officials resolving public procurement queries. Operating as a copilot without decision-making authority, it reduces legal risk while keeping final judgment in human hands.

The broader the AI's operational scope, the stricter state control must be to ensure final decision sovereignty remains with public officials.

Three Key Strategies for AI-Assisted Evaluation

To ensure this strict framework of requirements and accountability becomes a true engine of modernization rather than a bureaucratic barrier, technological implementation must center on clearly defined practical pillars. The effectiveness and legal certainty of AI proposal evaluation projects do not depend on improvisation, but on grounding deployment in three major strategic keys:

1. Process Sovereignty in Critical Functions: To ensure the long-term viability of these initiatives, public authorities must retain absolute control over both administrative procedures and strategic data. This requires prioritizing custom-designed evaluation systems so that the state maintains full ownership and governance over core management tools.
2. Active Governance & Continuous Oversight: Technological deployment does not end when an algorithm goes live in an operational environment. For the system to function effectively, it requires ongoing evaluation that combines rigorous pre-deployment testing with permanent human supervision, ensuring the machine always operates under correct criteria.
3. Proactive Transparency: AI tools must be designed to explain the rationale behind their scoring in an accessible, plain-language manner. Ultimately, both citizens and participating companies must clearly understand the grounds for every resolution in order to eliminate opacity and preserve trust in public institutions.

A "Yes" Conditioned on Excellence

Integrating AI into proposal evaluation is no longer an option, but an unavoidable necessity to prevent the Administration from collapsing under an exponential volume of data. However, this step does not represent a blank check, but a "yes" conditioned on operational excellence, as adopting this technology brings fundamental strategic value to the public sector at multiple levels.

First, artificial intelligence promotes enhanced procedural justice by eliminating errors caused by human fatigue and unconscious bias, ensuring that all proposals are evaluated using exactly identical parameters. Added to this is a crucial increase in public service agility, given that the drastic reduction in award times allows resources and aid to reach the citizens and institutions that need them much sooner.

On the other hand, the classification of this technology as "high risk" by European regulations should not be interpreted as an impediment, but as a genuine seal of quality. This classification aims to ensure that the Administration uses exclusively tools that are transparent, auditable, and equipped with contingency plans. As a result, this regulatory rigor ends up raising the standard of legal certainty far above traditional manual processes.

Ultimately, evaluation using artificial intelligence is the engine that will allow for the consolidation of a 21st-century Administration: faster, fairer, and more efficient. The real challenge is not technological, but one of governance. If the State takes the lead in the creation and oversight of these tools, AI will not replace civil servants, but will instead fortify their analytical capacity to make the best decisions in less time and for the benefit of all society.

‍

‍