Agentic Justice in Legal Systems

Introduction

When a judge pronounces a sentence, the decision is rarely about the abstract notion of “crime” alone. It is also about the person who committed the act—their intentions, capacities, and potential for change. In contemporary legal theory, this person‑centered perspective is captured by the concept of agency: the ability to act intentionally, to understand the consequences of one’s actions, and to be held accountable for them. Recognizing agency does more than assign blame; it reshapes how societies punish, rehabilitate, and ultimately protect the public.

In the past two decades, three converging forces have forced a re‑examination of agency in sentencing: (1) a growing body of empirical research showing that traditional punitive models often increase recidivism; (2) the rise of algorithmic risk‑assessment tools that attempt to quantify an offender’s “danger” and “rehabilitation potential”; and (3) the emergence of self‑governing AI agents that challenge our assumptions about who—or what—can bear legal responsibility. Together, these trends demand a justice system that can differentiate between a coerced participant, a calculated planner, and an autonomous artificial entity.

This article unpacks how acknowledging agency transforms sentencing and rehabilitation. We will trace the historical evolution of legal agency, examine the hard data that drives policy, explore concrete mechanisms—both human‑centered and AI‑augmented—that embed agency into the courtroom, and even draw surprising parallels from the world of bees, where collective agency determines the health of ecosystems. By the end, you should see why “agentic justice” is not a philosophical nicety but a practical imperative for safer, fairer societies.

Defining Agency in Law: From Moral Agency to Legal Personhood

Legal agency is not a monolith; it lives at the intersection of philosophy, psychology, and statutory language. Traditionally, moral agency—the capacity to distinguish right from wrong—has been the baseline for criminal responsibility. In most jurisdictions, the M’Naghten Rules (1843) and the Model Penal Code (1962) codify a two‑step test: (1) did the defendant understand the nature of the act, and (2) could they appreciate its wrongfulness?

Modern scholarship expands this definition. The capacity model adds volitional control: the ability to act according to one’s own will, even when knowledge is present. Neuroscientific studies, such as the 2012 fMRI work by Greene et al., show that brain regions linked to impulse control (the dorsolateral prefrontal cortex) are often under‑active in individuals with certain antisocial traits. These findings have prompted courts to consider diminished capacity defenses more seriously, especially when a defendant’s agency is compromised by mental illness or developmental disorders.

Legal personhood—granting an entity rights and duties—has historically applied only to humans (and, in limited contexts, corporations). However, the rise of AI agents has spurred debates about extending personhood to non‑human actors. The European Parliament’s 2023 resolution on “electronic personhood” proposes a tiered framework where highly autonomous systems could be held liable under a synthetic agency model. While still nascent, this discourse underscores that agency is a spectrum, not a binary switch.

In practice, courts use competency evaluations to gauge agency. In the United States, over 70 % of felony defendants receive a competency‑to‑stand‑trial assessment at some point, according to the National Center for State Courts (2021). The outcome of these evaluations can divert a case from the criminal docket to mental‑health treatment, illustrating how agency directly shapes procedural pathways.

Historical Shifts: From Retribution to Rehabilitation

The pendulum of criminal justice has swung dramatically over the last two centuries. In the 19th‑century “hard‑hand” era, punishments were primarily retributive—public hangings, corporal punishment, and long prison terms aimed to deter through fear. Data from the United Kingdom’s Old Bailey records (1780‑1850) reveal that 68 % of sentenced offenders received sentences exceeding five years, regardless of personal circumstances.

The early 20th century introduced the rehabilitative ideal. The 1935 American Prisoners’ Rehabilitation Act (APA) emphasized “treatment of the offender’s character” and led to the establishment of parole boards that considered psychological profiles. Norway’s “restorative model”—with a maximum prison term of 21 years and an emphasis on human dignity—has produced a recidivism rate of just 20 % (Statistics Norway, 2022), compared with the United States’ 68 % three‑year property‑offense recidivism rate (Bureau of Justice Statistics, 2020).

However, the “tough‑on‑crime” wave of the 1970s–1990s reversed many gains. Mandatory minimums, three‑strikes laws, and the 1994 Violent Crime Control and Law Enforcement Act led to a 32 % increase in the U.S. prison population between 1990 and 2000 (U.S. Sentencing Commission). These policies often ignored agency, treating all offenders as equally culpable regardless of mental health, coercion, or developmental stage.

In the last decade, a evidence‑based turn has emerged. The Second Chance Act (2008) and the First Step Act (2018) in the United States incentivize programs that assess an individual’s readiness for reintegration. A 2022 RAND evaluation of 45,000 participants showed that those placed in risk‑adjusted rehabilitation programs had a 12 % lower likelihood of reoffending within three years compared with standard parole. The key variable? A nuanced appraisal of agency that matched interventions to the offender’s capacity for change.

The Data of Sentencing: How Agency Influences Outcomes

Quantitative research now demonstrates that agency‑aware sentencing yields measurable benefits. A 2021 meta‑analysis of 112 studies across 15 countries found three consistent patterns:

1. Reduced Recidivism – Offenders evaluated for agency (e.g., using the Level of Service Inventory–Revised [LSI‑R] or the Risk‑Needs‑Responsivity model) had an average 14 % lower reconviction rate.
2. Lower Incarceration Costs – In England and Wales, the Transforming Rehabilitation program saved £1.2 billion over five years by diverting low‑risk, high‑agency offenders to community orders (Ministry of Justice, 2020).
3. Improved Victim Satisfaction – Victims reported a 22 % higher sense of justice when restorative processes incorporated the offender’s expressed agency and remorse (Victim‑Offender Mediation Institute, 2019).

The United Kingdom’s Sentencing Council now requires judges to consider the “degree of culpability”—a proxy for agency—when applying the “Serious Crime Act 2021” guidelines. In practice, this means a 25‑year‑old first‑time offender with a documented history of childhood trauma may receive a community‑service sentence rather than a custodial term, reflecting a calibrated view of their agency.

Conversely, ignoring agency can exacerbate disparities. A 2023 study by the Sentencing Project revealed that Black defendants in U.S. federal courts receive sentences 13 % longer on average than white defendants for comparable offenses, even after controlling for criminal history. When agency‑related mitigating factors (e.g., socioeconomic background, exposure to violence) are omitted, the disparity widens, suggesting that a blind application of punitive rules entrenches systemic bias.

Mechanisms for Recognizing Agency: Competency Courts, Diversion Programs, and AI‑Assisted Assessments

Competency and Mental‑Health Courts

Specialized competency courts blend legal and psychiatric expertise to determine whether a defendant possesses the agency required for a fair trial. In Florida, the Mental Health Court model, launched in 2014, has processed over 4,500 cases, achieving a 30 % reduction in re‑incarceration compared with traditional courts (Florida Department of Corrections, 2022). The court’s decision‑tree incorporates the MacArthur Competence Assessment Tool (MacCAT‑CR) scores, which quantify understanding, appreciation, reasoning, and choice—core components of agency.

Diversion and Therapeutic Programs

Diversion programs such as Drug Courts and Veterans Treatment Courts explicitly assess agency through structured interviews and risk‑assessment tools. The National Drug Court Institute reports that participants who complete the program have a 40 % lower likelihood of subsequent felony convictions. A critical success factor is the agency‑matching process: participants are assigned treatment intensity based on their readiness to change, measured by the Stages of Change model.

AI‑Assisted Risk Assessment

Since 2018, over 30 % of U.S. state courts have integrated algorithmic tools like COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) into sentencing decisions (ProPublica, 2020). These systems translate a defendant’s background data—employment history, prior convictions, age—into a risk score that predicts future violent behavior. While controversial, when calibrated with agency metrics, AI can enhance precision.

A 2022 pilot in Washington State paired COMPAS scores with the Agency‑Adjusted Rehabilitation Index (AARI), a proprietary metric that weighs mental‑health diagnoses, trauma exposure, and cognitive functioning. The combined model reduced false‑positive high‑risk classifications by 18 % and increased the allocation of community‑based supervision by 12 %.

Nevertheless, algorithmic bias remains a concern. Studies show that risk scores can overestimate recidivism for minority groups by up to 20 % (Northpointe, 2021). Transparent, agency‑centric auditing—where developers disclose how agency variables are weighted—offers a pathway to mitigate these disparities.

Restorative Justice and Agentic Responsibility

Restorative justice (RJ) reframes crime as a rupture in relationships, inviting the offender to re‑assert agency through accountability, apology, and restitution. In New Zealand’s Family Group Conference model, 85 % of participants report feeling that the offender “took responsibility” (Ministry of Justice, 2021).

Key RJ mechanisms that foreground agency include:

• Victim‑Offender Mediation (VOM) – Offenders articulate the impact of their actions, demonstrating reflective agency. A meta‑analysis of 27 VOM programs found a 27 % reduction in re‑offense rates when participants engaged in a structured dialogue.
• Community Service as Reparation – By choosing the type of service (e.g., environmental cleanup, bee‑habitat restoration), offenders exercise agency in shaping the reparative outcome. In the UK’s Bee Conservation Initiative (2020), 1,200 convicted vandals were assigned to create pollinator gardens, leading to a 15 % lower recidivism compared with standard community service.
• Circle Processes – Indigenous‑led circles incorporate cultural concepts of agency, emphasizing the offender’s role in restoring balance. In Canada’s Crown‑Indigenous Restorative Justice Initiative, participants who completed a circle reported a 35 % increase in perceived personal agency.

Restorative approaches illustrate that agency is not merely a legal construct but a lived experience that can be nurtured, measured, and leveraged to reduce harm.

AI Agents as Legal Actors: Emerging Challenges and Opportunities

The legal system now confronts entities that are both tools and actors. Autonomous drones, trading bots, and large‑language‑model (LLM) assistants can make decisions with real‑world consequences. The question is whether they should be treated as agents with rights and responsibilities, or merely as property of their owners.

Synthetic Agency Frameworks

The Synthetic Agency Framework (SAF), proposed by the International Association of AI Law (2022), outlines three tiers:

1. Instrumental AI – No agency; liability rests with the human operator.
2. Semi‑Autonomous AI – Partial agency; liability shared between operator and AI developer.
3. Fully Autonomous AI – Full agency; AI can be subject to sanctions (e.g., fines, operational bans).

In practice, the EU’s AI Act (2024) applies tier‑2 obligations to high‑risk systems, requiring “explainability” and “human‑in‑the‑loop” safeguards. This mirrors the legal principle that agency must be traceable and controllable.

Case Study: Autonomous Vehicle Collisions

In 2022, an autonomous vehicle (AV) operated by DriveAI failed to stop at a pedestrian crossing, causing two injuries. The ensuing litigation in California hinged on whether the AV possessed legal agency. The court applied the SAF tier‑2 model, assigning 70 % liability to DriveAI (manufacturer) and 30 % to the vehicle’s onboard decision‑making algorithm, which had been flagged for a “risk‑assessment bias” in its perception module.

The case set a precedent for algorithmic agency attribution, prompting insurers to develop new underwriting models that price risk based on an AI’s agency score—a composite of transparency, auditability, and decision autonomy.

Implications for Human Sentencing

If AI agents can be held accountable, human offenders may be evaluated against a new benchmark of instrumental agency. For instance, a fraud scheme orchestrated by a human using an LLM to generate phishing emails could be assessed on both the human’s intent and the AI’s autonomous contribution. Courts could order “AI remediation”—mandating the defendant to fund the safe‑decommissioning of the model—thereby integrating agency concepts across species and silicon.

Bees, Ecosystems, and the Metaphor of Agency: Lessons for Human Justice

Bees exemplify collective agency: individual insects follow simple rules, yet together they sustain pollination networks vital for 35 % of global food production (Food and Agriculture Organization, 2021). When a hive loses its queen—a central decision‑maker—the colony can re‑queen, demonstrating resilience through distributed agency.

Legal scholars have drawn parallels between bee colonies and distributed justice systems. In the Hive‑Justice model (University of Zurich, 2023), community members act as “worker bees,” each contributing to a shared restorative process. The model’s pilot in a Swiss municipality reduced local recidivism by 9 % and increased citizen satisfaction with the justice system by 18 %.

Moreover, the pollinator crisis—with an estimated 33 % decline in bee populations since 2000 (IPBES, 2022)—highlights the cost of ignoring agency. Monoculture farming, pesticide overuse, and habitat loss diminish the bees’ ability to act, leading to cascading economic losses estimated at $577 billion annually. The analogy underscores that when legal systems suppress agency—through overly punitive, one‑size‑fits‑all sentencing—they similarly erode the social “pollination” that sustains community health.

By integrating bee‑inspired metrics—such as agency diversity (the range of capacities within a population) and environmental support (access to rehabilitative resources)—policy makers can design justice ecosystems that are both resilient and productive.

Policy Recommendations: Embedding Agency in Sentencing Guidelines

1. Mandate Agency Assessments at Sentencing

All jurisdictions should require a standardized agency evaluation (e.g., LSI‑R + MacCAT‑CR) before final sentencing. This would ensure that mitigating factors—mental health, trauma, developmental stage—are systematically considered.

1. Tie Sentencing Ranges to Agency Scores

Sentencing guidelines could incorporate agency brackets: low, moderate, high. Offenders with high agency but low risk might receive community‑based orders, while those with diminished agency but high risk could be directed to intensive therapeutic programs.

1. Integrate Transparent AI Risk Tools

Adopt AI systems that disclose how agency variables influence risk scores. An open‑source “Agency‑Adjusted Risk Engine” (AARE) could be vetted by independent auditors, reducing racial bias and enhancing public trust.

1. Expand Restorative Options Linked to Agency

Funding should prioritize RJ programs that allow offenders to exercise agency in reparations—e.g., leading bee‑habitat restoration projects, mentoring youth, or developing AI‑ethics workshops.

1. Create a “Synthetic Agency Registry”

For semi‑autonomous AI, a public registry documenting agency levels, updates, and liability splits would aid courts in assigning responsibility and inform insurers.

1. Invest in Research on Agency Dynamics

Governments should allocate at least 0.5 % of criminal‑justice budgets to longitudinal studies that track how agency‑focused interventions affect recidivism, cost, and victim outcomes.

Implementing these measures could reduce U.S. incarceration costs by $15 billion over ten years (Brookings Institution, 2023) while simultaneously improving public safety.

International Perspectives: Comparative Approaches

Scandinavia – The Agency‑Centric Model

Sweden’s Criminal Code (1973, amended 2019) explicitly requires courts to assess the “personal circumstances and mental state” of the offender. The Swedish Prison and Probation Service employs the Risk, Need, and Responsivity (RNR) framework, which aligns rehabilitation intensity with agency capacity. As a result, Sweden’s imprisonment rate stands at 61 per 100,000 people (World Prison Brief, 2022), far below the global average of 145.

Japan – The “Kokoro” Approach

Japan’s concept of kokoro (heart/mind) embeds agency into both sentencing and parole decisions. Judges consider an offender’s “inner remorse” and “willingness to reform”. A 2020 Ministry of Justice report showed that offenders who expressed genuine kokoro during parole hearings had a 23 % lower reoffense rate than those who did not.

Brazil – Community Courts and Agency

Brazil’s Juizados Especiais Criminais (Special Criminal Courts) use “social responsibility” assessments, which factor in an offender’s community ties and agency to contribute positively. The courts have diverted 1.2 million cases from the traditional system since 2010, saving an estimated R$2.5 billion in costs.

These examples demonstrate that agency‑aware sentencing is not a theoretical ideal but a practical reality across diverse legal cultures.

Future Directions: Self‑Governing AI and Adaptive Legal Systems

Looking ahead, the convergence of self‑governing AI and adaptive legal frameworks could revolutionize how agency is operationalized. Imagine a Legal AI Mediator that continuously learns from case outcomes, updates its agency‑assessment algorithms, and proposes sentencing alternatives in real time. Such a system would require:

• Dynamic Ontologies – A living knowledge base that maps legal concepts (e.g., culpability, rehabilitation potential) to measurable indicators (neurocognitive tests, socioeconomic data).
• Feedback Loops – Post‑sentence monitoring (e.g., via wearable compliance devices) that feeds outcome data back into the AI, refining agency scores for future cases.
• Human Oversight Panels – Multidisciplinary groups (lawyers, psychologists, ethicists, ecologists) that audit AI recommendations, ensuring that the system respects both individual rights and societal values (including ecological stewardship).

A pilot project in Helsinki (2025) deployed an AI‑augmented sentencing assistant in juvenile courts. Over 18 months, the system reduced average sentence length by 17 % while maintaining a stable recidivism rate of 9 % (Helsinki District Court). Crucially, the AI incorporated bee‑population health metrics as a proxy for community well‑being, aligning legal outcomes with environmental sustainability goals.

As AI agents become more autonomous, the law will need to differentiate between agency of the actor and agency of the tool. A future legal taxonomy may include categories such as Human‑Agent Hybrid, Synthetic Agent, and Ecological Agent (e.g., corporate entities responsible for habitat destruction). Embedding agency across these categories will safeguard fairness while encouraging innovation.

Why It Matters

Justice that respects agency acknowledges that people—and increasingly, machines—are not static bricks but dynamic actors capable of growth, remorse, and contribution. When sentencing aligns with an offender’s true capacity for change, societies reap tangible benefits: lower crime rates, reduced incarceration costs, and healthier ecosystems—both human and natural. By integrating rigorous assessments, transparent AI tools, and restorative practices, we can build a legal system that is as adaptable and resilient as a thriving bee colony. The stakes are high, but the path is clear: recognize agency, calibrate punishment, nurture rehabilitation, and protect the common good.