THE LONG VIEW

Bottom Line Up Front: The absence of benchmarks for doctrinal legal reasoning in AI systems creates a critical compliance and operational risk under the EU AI Act, threatening the legitimacy of automated legal decision-making. Threat Identification: Current legal AI evaluation frameworks fail to assess doctrinal reasoning—the interpretive, precedent-based analysis central to legal practice—focusing instead on superficial or paralegal tasks. This creates a 'measurement gap' that prevents meaningful enforcement of the EU AI Act’s 'appropriate accuracy' requirement for high-risk legal AI systems [Finck, 2026]. Probability Assessment: The threat is already materializing. As of 2026, AI tools are increasingly deployed in judicial and legal advisory roles across the EU. Without a benchmark, compliance with the AI Act’s accuracy mandates remains unverifiable. Full-scale systemic risk is highly likely within 1–3 years as adoption accelerates. Impact Analysis: The consequences include legally unsound decisions, diminished judicial accountability, and potential violations of the right to a fair trial. Regulatory enforcement becomes ineffective without measurable standards, undermining public trust in both AI and legal institutions. High-risk domains such as asylum, criminal sentencing, and social benefits are especially vulnerable. Recommended Actions: 1) Fund interdisciplinary development of doctrinal reasoning benchmarks involving legal scholars and AI researchers; 2) Establish a European Legal AI Evaluation Authority to certify high-risk systems; 3) Issue EU Commission guidance clarifying that 'appropriate accuracy' includes interpretive fidelity to legal doctrine; 4) Mandate transparency in legal AI training data and reasoning pathways. Confidence Matrix: Threat Identification – High confidence (based on documented limitations of current benchmarks); Probability Assessment – Medium-High confidence (inferred from deployment trends and regulatory timelines); Impact Analysis – High confidence (rooted in legal principle and precedents on judicial accountability); Recommended Actions – Medium confidence (dependent on political and institutional will).