KÜHEYLAN HUKUK
The digital age we are in has led to a radical change in criminal law and security policies by increasing the processing capacity of huge data masses. The sanctioning character of traditional criminal law, which focuses on the aftermath of the crime, is being replaced by a proactive and predictive structure that targets the pre-crime, with the opportunities offered by big data and artificial intelligence technologies. As a reflection of this transformation, predictive policing systems adopted by law enforcement units around the world today; It aims to predict with statistical probabilities where a crime may be committed or who has the potential to commit a crime by processing large data sets such as past crime data, geographical location information, facial recognition records and social media movements through algorithms.
In this study, the legal nature of artificial intelligence-supported crime prediction systems is analyzed through current legal/institutional application examples in the world and in Türkiye, and the structural problems created by these systems in the context of the preventive function of criminal law are discussed. In the study; It is argued that crime prediction algorithms transform the concept of reasonable suspicion, which is one of the basic building blocks of criminal procedure, into algorithmic suspicion, separate individuals from the concrete plot, reduce them to a risk score, and classify them as potential criminals in advance. In addition, it has been determined that the self-censorship mechanism that emerged as a result of the deterrent effect on individuals created by the uninterrupted surveillance infrastructure built on a social scale, eroded fundamental democratic rights such as freedom of expression and the right to assembly and demonstration. In conclusion, it is emphasized that the criminal justice process cannot be reduced to pure statistical calculations; While using artificial intelligence in the fight against crime, it has been revealed within legal and ethical boundaries that the balance of freedom and security should be maintained in favor of the presumption of innocence, the principle of fault and the guarantees of the rule of law.
Function, Algorithmic Suspicion, Presumption of Innocence, Chilling Effect.
The age of digitalization we are in has not only transformed every aspect of social life, but also deeply affects security and criminal justice policies, which are one of the most basic duties of the state. Traditional criminal law, as a rule, has a punitive character that comes into play after a legal value is violated, that is, after the crime is committed or at least after enforcement actions begin. Today, the analytical capacity reached by big data and artificial intelligence (AI) technologies is shifting the focus of the criminal justice system from post-crime intervention to pre-crime prediction and prevention.
Predictive policing systems, which are increasingly used by law enforcement agencies around the world, aim to predict where a crime may be committed or which individual has the potential to commit a crime by using statistical algorithms by processing huge data sets. Although these technological possibilities offer the promise of perfect security at first glance, in the context of the state's obligation to maintain public order and prevent crimes, they erode gains such as the principle of fault, causal link and presumption of innocence, which form the basis of criminal law.
Crime prediction applications change the relationship between the state and the citizen. Artificial intelligence systems remove the law enforcement officer's discretion and the physical and informational context in which he evaluates events; The concept of concrete and reasonable suspicion in criminal procedure is replaced by a statistical algorithmic suspicion. These systems, which work with correlations that are mostly reflections of past police practices rather than the reasons that give rise to crime, separate individuals from the plot, reduce them to a risk score and classify them as potential criminals.
This study consists of three main parts. In the first part, the concept of artificial intelligence, its historical development, capacity levels and its legal nature from the perspective of the Turkish Civil Code and Administrative Law will be discussed. In the second part, the theoretical foundations of criminal law, general and special prevention theories, the risk society phenomenon and the conceptual framework of predictive policing will be discussed. In the third part, current artificial intelligence systems used in the fight against crime will be examined through international (HART, COMPAS) and national (PTS, KGYS) examples; Violations of these systems on criminal law principles and fundamental rights and freedoms, ethical-legal dilemmas, the need for regulation on the axis of the EU Artificial Intelligence Act (AI Act) and solution proposals will be evaluated with a holistic approach. In the conclusion section, a general evaluation and legislative recommendations will be presented in the light of the findings.
The concept of artificial intelligence (AI) was used for the first time in 1955 by mathematician John McCarthy in the scholarship program he applied for to fund his computer project within the framework of the Dartmouth Conference. .
Although artificial intelligence is not described in a standard way, it is generally defined in the doctrine as systems that can use human-specific features such as learning, interpretation and decision-making3.
According to the Council of Europe Framework Convention on Artificial Intelligence, artificial intelligence is software that prepares outputs such as content, suggestions and predictions for purposes determined by humans, using methods such as induction, deep learning and statistical approach, and can make decisions according to their context. In the text submitted to the commission in 2023, artificial intelligence was described as "a machine-centered system produced for the purpose of working autonomously, characterizing physical and hypothetical environments, producing suggestions and decisions, in line with specified purposes"7.
Artificial intelligence according to the 3 A model proposed by Yanisky-Ravid and Liu; They are systems that can make decisions without external intervention, update themselves by learning from data, and produce rational outputs. Characteristic features of these systems, such as autonomy and unpredictability, take artificial intelligence beyond being a mere technical tool and necessitate a re-evaluation of its legal status and liability regime 8 .
It is possible to divide artificial intelligence into three different types: inspired, humanoid and analytical. Analytical artificial intelligence is a type of artificial intelligence that can use previously used data to make future decisions by learning with the machine learning method. Autonomous structures that can include emotional intelligence (EQ) in decision-making processes in addition to cognitive capacity (IQ) are described as human-inspired artificial intelligence. Humanoid artificial intelligence, which represents a more advanced stage of this category; It refers to systems that have the ability to interact autonomously with their environment and adapt to social dynamics by adding a social dimension to their cognitive and emotional competencies 11 .
Artificial intelligence systems in the literature; They are subject to a triple classification as narrow, general and super artificial intelligence, according to their functional capacities and competence levels. Narrow artificial intelligence, which constitutes the first step of this classification, refers to systems that are designed to perform only predefined specific tasks and display expertise limited to this field. General artificial intelligence, which aims to model the reasoning and reasoning processes of the human mind, is a category that is still in the development phase and has the potential to engage in versatile cognitive activities. Finally, super artificial intelligence; It represents a speculative species that exceeds the capacity of human intelligence, can manage complex operations simultaneously, and is theoretically predicted to go beyond human will. Today, this category is discussed mostly in the context of futuristic approaches and dystopian narratives.
It is possible to classify artificial intelligence into four types. The first type of artificial intelligence is the robotic type, which can make and implement decisions on limited issues with its past knowledge and programmed abilities, the second one can react to the environment and shows expertise in a single field, and the third type has broader capabilities. The fourth and last type is the type that can be called super artificial intelligence, which is consciously aware of itself and can explain its feelings and thoughts to its surroundings, but has not been produced yet. It is also possible to express these types technically as reactive machines, limited memory, theory of mind and self-awareness.
These theoretical classifications in the doctrine draw the technical boundaries of today's predictive policing practices. All algorithmic systems currently used operationally by law enforcement are based on structurally narrow AI and functionally on analytical AI and limited memory models. This indicates that these security systems do not have a general cognitive capacity or human self-awareness; On the contrary, it shows that there are only specific software focused on a specific purpose. Using machine learning and data analytics methods, the system finds patterns through past crime records, geographical coordinates and KGYS integrations, and transforms this historical data in its limited memory into a statistical probability output for the future. Therefore, the main source of legal problems encountered in crime prediction is not because artificial intelligence has reached a futuristic consciousness; It arises from the fact that the outputs produced by pure statistical correlations in a narrow field of expertise directly conflict with the concepts of criminal law such as perpetrator, fault and causality.
According to the European Union Artificial Intelligence Law, which is the world's first regulation on artificial intelligence, artificial intelligence systems are divided into types based on risk. According to this law, artificial intelligence systems are classified as prohibited, high risk, limited risk and minimized risk. Through this classification, the European Union aimed to ban systems that pose unacceptable risks in terms of fundamental rights and freedoms; It also aims to ensure legal predictability by clarifying the liability regime of stakeholders in the artificial intelligence ecosystem.
There is no legal regulation regarding artificial intelligence technologies in the Turkish legal system yet. While the status of artificial intelligence remains unclear within the framework of the legislation, different approaches to the legal nature of these systems are exhibited in the doctrine. In this context, artificial intelligence; It is being discussed whether it will remain in its current status as an item or be subjected to a regime similar to slavery law. On the other hand, the views that the legal entity model should be applied for developed autonomous structures or that a sui generis legal status should be created under the name of electronic personality constitute the main discussion axes in the doctrine14.
In the 8th and subsequent articles of the Turkish Civil Code No. 4721, real persons are regulated as subjects of rights; Articles 47 and following include provisions regarding legal entities15. In our legal system, real personality refers to a status reserved only for humans; Legal entity defines groups of persons or property that have a legal existence independent of the will that constitutes them. Within this traditional distinction, artificial intelligence can neither be included in the category of a natural person as a biological entity nor as a legal entity that meets current legal conditions16. The legal entity institution has a deep legal history, with its historical roots dating back to Roman Law. In the face of the limitation of human life, the concept of legal personality was created in order to ensure the continuity of certain aims to be achieved and to make these aims independent of the existence of individuals. In this way, the continuity of collective goals that exceed the lifespan of natural persons has been made possible at the legal level through an independent subject of rights17. Views centering on the autonomous characteristics of artificial intelligence categorically reject the characterization of these systems as mere goods. According to this approach; Although it is clear that artificial intelligence does not have a biological reality, the high cognitive competencies and reasoning potential it exhibits constitute a rational basis for granting it a status similar to the legal entity model. Therefore, in the doctrine, the independence of artificial intelligence in decision-making mechanisms constitutes the main basis for claims that it should be positioned beyond traditional property law in the hierarchy of legal subjects18.
The Robotics Recommendation Report was published by the European Parliament in 2017. According to this report, an electronic personality approach has been introduced to these systems. In the draft framework agreement accepted on February 12, 2024, artificial intelligence is a machine-based network that predicts how results such as suggestions, content, assumptions or decisions that may affect will be created from the inputs it receives, for explicit or implicit goals. The electronic personality approach has been abandoned for now, stating that different artificial intelligence systems differ in their levels of autonomy and adaptability after deployment20.
Goods are assets that have material existence and can be dominated over, have economic value and do not have personal value. In accordance with the property right regulated in Article 683 of the Turkish Civil Code No. 4721, the owner; It has the authority to use (usus), benefit (fructus) and dispose (abusus) of goods within the boundaries of the legal order. Although artificial intelligence systems are described as goods subject to this property regime in today's legal practice, the autonomous decision-making abilities and unpredictable behavior patterns of these systems result in the classical property law norms being inadequate in terms of liability regime. In this context, accepting artificial intelligence as a mere object creates a doctrinal gap, especially in terms of compensating the damages caused by these systems and sharing legal liability.
Considering the level that artificial intelligence has reached, both in terms of administrative law and private law, it is an appropriate approach to evaluate these systems as goods at the current stage. In accordance with Article 128 of the Constitution, it is essential that the essential and permanent duties required by public services be carried out only by civil servants and other public officials. In this context, artificial intelligence systems that do not have real personality and therefore do not have Turkish citizenship status; It is not legally possible to qualify as a public servant, worker or auxiliary personnel.
However, it is an undeniable fact that legal norms lag behind technological developments and this creates a regulatory gap. Although it is foreseen in future projections that national legal systems may grant unique statuses to artificial intelligence, such as legal personality or electronic personality, the situation is different when an evaluation is made within the scope of the current legislation and the current technological level. Whether used by the administration in independent decision-making processes or through a public official as a decision support system; Artificial intelligence technology does not have an independent legal personality today, but should be considered an advanced technological tool used in the transactions and actions of the administration.
The most basic duty of the administrative police is; To ensure that individuals can exercise their constitutional rights and freedoms on a safe basis by establishing public order and security. Developing artificial intelligence technologies offer remarkable operational opportunities to the administration in the execution of this protective and preventive function of the law enforcement. In this regard, algorithmic systems are increasingly being used by the security bureaucracy. Artificial intelligence applications greatly expand the intervention capacity and sphere of influence of the law enforcement by providing speed, efficiency and resource optimization in the process of detecting and preventing possible crimes in advance. Ultimately, this technological integration aims to minimize crime rates and ensure public safety. In parallel with this security paradigm change on a global scale, the use of artificial intelligence-supported analytical systems in crime fighting strategies is becoming increasingly widespread in Türkiye.
Artificial intelligence technologies are developing with great acceleration and new artificial intelligence systems with different qualities are being put into service every day. These systems; It can analyze and evaluate personal data that is already in the database, later included in the system or obtained from accessible sources, in line with the specified purposes. Moreover, artificial intelligence applications not only process existing data; It can also be used to obtain new information about a targeted person by analyzing this data.
Artificial intelligence systems; Thanks to features such as movement, voice or DNA analysis, it is possible to detect individuals who are suspicious or have the potential to commit a crime by looking at their posture or movements; It also provides the opportunity to identify inconsistencies or untrue statements in discourse and expressions. Similarly, through facial recognition systems, the places visited by the person are determined based on the data in the records, and certain profiles can be accessed by behavioral analysis through artificial intelligence. Artificial intelligence allows large amounts of data to be analyzed instantly for specific purposes. For example; By analyzing the places of worship such as mosques, churches or synagogues visited by people, their clothing styles and social behaviors by artificial intelligence, the person's estimated crime potential can be determined and preventive measures can be taken. However, it should not be ignored that these algorithmic evaluations may not always produce accurate results and that they carry the risk of violating people's fundamental rights and freedoms.
Thanks to technological possibilities and artificial intelligence; By analyzing Big Data available on platforms such as image and facial recognition systems and search engines, the risks of individuals committing crimes for the first time or repeatedly can be evaluated. This situation allows protective and preventive measures to be taken before a crime is committed. Therefore, artificial intelligence technologies; It offers unprecedented operational opportunities to law enforcement forces in the processes of preventing, monitoring and illuminating crime. As a matter of fact, within the context of the new security strategy adopted in the European Union in 2015, through the CoESS and UNI projects implemented with the financial support of the European Commission; It is aimed to develop and use systems that integrate human skills and high technology in a wide range of security issues ranging from financial crimes to terrorist acts.
When the historical development of criminal law is examined, it is seen that the classical understanding of law, as a rule, comes into play after the crime is committed and focuses on punishing the perpetrator. Modern criminal law abandons the idea of revenge; It has undertaken a preventive duty aimed at protecting public order and preventing crime before it is committed. Article 1 of the Turkish Penal Code No. 5237 clearly states that the purpose of the criminal law is "to prevent the commission of a crime", thus giving a legal basis to this protective function24.
Preventive function of criminal law in doctrine; It is evaluated on two main axes: general prevention and special prevention. General prevention refers to the fact that the sanctions prescribed by law prevent potential perpetrators from committing crimes by creating a psychological pressure and deterrence on the general society26. Special prevention, on the other hand, aims to rehabilitate a particular individual who has previously committed a crime or who has a tendency to commit a crime and to reintegrate him into society, thus preventing him from committing a crime again27.
Today, the transformation of post-industrial societies into a risk society, as defined by Ulrich Beck, has increased the need for security to the highest level and made the preventive role of the criminal justice system even more important28. In the modern state understanding, individuals, rather than punishing the perpetrator after the danger occurs and the damage occurs; It demands that the danger be detected and eliminated while it is still at a risk stage. This expectation forces criminal law and law enforcement activities to evolve from a reactive structure to a predictive structure.
The use of algorithmic systems and artificial intelligence by security units is a reflection of this desire to take action in advance. The fact that law enforcement produces mathematical probabilities about crimes that have not yet been committed by analyzing personal data appears as the most concrete and at the same time the most controversial view of the preventive function of criminal law in the digital age in terms of fundamental rights and freedoms.
This digital transformation leads to a qualitative change in the concept of suspicion, which is one of the most basic building blocks of criminal procedure law. In the traditional criminal justice system, the implementation of judicial law enforcement measures is based on concrete facts reflected in the outside world and a process based on human reason. However, artificial intelligence-supported crime prediction algorithms reduce suspicion to an inductive statistical probability rather than a concrete factual basis. This technological approach, which ignores the individual's freedom of will and capacity to make decisions for the future, has the potential to shift the focus of criminal law from the concrete act that has occurred to the subject who is likely to commit a crime in the future, thus creating a situation that pushes the limits of the presumption of innocence.
On the other hand, automating the preventive function through algorithms also blurs the historical and structural boundaries between the administrative and judicial duties of law enforcement forces. In accordance with the doctrine, while administrative law enforcement acts against general dangers in order to prevent the commission of a crime; Judicial law enforcement exhibits a reactive nature in line with the orders of the judicial authorities in order to reveal the material truth after the crime is committed. Artificial intelligence's individual and pinpoint risk scoring before a criminal act occurs brings administrative law enforcement activities to the intensity of intervention of the judicial police, but this process remains outside the judicial control mechanisms and judge guarantees arising from criminal law. Therefore, maintaining the balance between the protective function of criminal law and the protection of fundamental rights requires this new predictive law enforcement model to be regulated within constitutional limits.
Artificial intelligence-supported predictive policing applications should be considered as the most concrete tools adapted to the digital age for the general and specific prevention purposes included in the classical theories of criminal law. The general prevention function, traditionally carried out through the abstract deterrence of penalties in the law, is turning into a dynamic control mechanism on potential perpetrators, thanks to the uninterrupted surveillance and location-oriented risk predictions offered by artificial intelligence. This algorithmic positioning, which increases the visibility of law enforcement in areas where crime will be concentrated, reinforces the deterrence purpose of general prevention at a practical level by concretizing the risk of being caught for individuals determined to commit crimes.
On the other hand, person-oriented algorithmic modeling directly interferes with the limits of private prevention theory. Risk scoring, which is carried out based on individuals' past actions and socio-demographic data, aims to prevent people with a high tendency to commit crimes from committing crimes by tracking them before a violation occurs. However, this situation, combined with the desire to eliminate the danger brought by the risk society at its source, overshadows the philosophy of rehabilitation and reintegration, which is the traditional aim of private prevention. The system turns the individual from being a rational and free-willed subject into an object of risk that must be kept under control.
This situation points to a serious change in mentality in the institutional functioning of law enforcement agencies. When we look at current law enforcement practices in Continental Europe, we see that these artificial intelligence-supported technologies are found in a wide range of areas, from the fight against terrorism to border security, from the protection of public order to organized crime analysis. While this technological integration promises operational success in reducing crime rates, on the other hand, it creates serious gray areas on the basis of fundamental rights and freedoms such as the presumption of innocence, prohibition of discrimination and protection of personal data.
One of the issues where the mentioned gray areas become most evident is the inherent bias of algorithmic systems and the risk of structural discrimination that arises accordingly. Artificial intelligence models, by their nature, are fed with empirical data from past periods and rationalize human or institutional tendencies in these data and carry them to the future. When data from socio-economically disadvantaged regions or certain social groups, where law enforcement agencies have historically exercised more intense control, is uploaded to the system, the algorithm codes these past practices as an absolute crime risk and recommends greater police visibility in those areas. This situation has the potential to systematically undermine equality before the law and the prohibition of discrimination, which are universal principles of criminal law, by creating a vicious circle called self-fulfilling prophecy.
However, the most radical break caused by predictive policing practices in terms of criminal procedure law is the deterioration of the standards of suspicion, which is a prerequisite for protective measures. In classical reasoning, in order to resort to judicial law enforcement measures that restrict fundamental rights such as search, arrest or detention, the existence of a standard of suspicion reflected in the outside world, auditable and based on concrete facts is required. However, the risk scores produced by artificial intelligence, their source and data weighting mechanisms are like a black box that cannot be known by the defense authority. Triggering judicial protection measures based on the abstract output of a system whose algorithmic logic cannot be audited on the grounds of trade secret or corporate secrecy poses the danger of rendering dysfunctional the principles of equality of arms and adversarial trial, which are the founding elements of the right to a fair trial.
Artificial intelligence-supported crime prediction systems have turned into tools integrated into the operational processes of law enforcement forces on a global scale. These systems are classified into two main categories, based on their architectural designs and focus points: place-based prediction models, which are based on geographical coordinates, and person-based prediction models, which directly process individuals' data.
One of the most concrete examples of person-based predictive policing activities internationally is the HART (Harm Assessment Risk Tool) system developed by Durham Police in the United Kingdom. The HART system is a decision support mechanism designed to predict the likelihood of suspects re-offending. The system classifies individuals as low, medium or high risk by analyzing 34 different data categories such as age, gender, past criminal records and zip code containing socio-demographic characteristics. At the end of a two-year trial period, it was determined that the system achieved a statistical success of 98% in predicting individuals as "low risk" and 88% in predicting individuals as "high risk". Law enforcement and judicial authorities use this algorithmic scoring as a reference point in the parole, bail or probation processes of suspects.
COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) software, which is widely used in the United States judicial system, creates risk profiles of defendants at different stages of criminal execution and trial. Similarly, the SSL (Strategic Subject List) system used by the Chicago Police Department lists people most at risk of being involved in or victimized by gun violence in the city.
One of the most radical examples of the integration of physical surveillance and artificial intelligence is the Integrated Security Camera System implemented throughout Moscow by the Ministry of Internal Affairs of the Russian Federation. This smart video analytics network, which has been expanded since 2019, can instantly scan the biometric facial data of people in controlled areas and match it in a central database of 500 million people in as little as 0.3 seconds. Through this system, which also allows the measurement of social and demographic characteristics of the population, data on wanted persons is transmitted to law enforcement units in real time. During the pandemic in 2020, more than 200 quarantine violators were detected in a single day only through facial recognition algorithms and were subject to administrative sanctions. Similarly, in the financial field, artificial intelligence-based systems such as QuantaVerse provide protection by analyzing complex account movements to prevent white-collar crimes such as money laundering (AML) and fraud.
When the technological and legislative infrastructure of predictive and preventive law enforcement activities in the context of Türkiye is examined, it is seen that the crime prevention duty of administrative law enforcement is supported by a wide digital network. At the legislative level, Additional Article 7 of the Police Duties and Powers Law (PVSK) No. 2559 and Additional Article 5 of the Gendarmerie Organization, Duties and Powers Law grant law enforcement units the authority to collect preventive intelligence in order to protect the indivisible integrity of the state with its territory and nation, constitutional order and general security. The License Plate Recognition System (PTS), which works integrated with the City Security Management System (KGYS) established within the framework of this legal authority, provides raw data to artificial intelligence analytics by instantly processing vehicle movements, geographical routes and license plate characters throughout the country.
The "Department of Big Data and Artificial Intelligence Applications" established within the Digital Transformation Office of the Presidency of the Republic of Turkey carries out the strategic planning of artificial intelligence systems in the public sphere and therefore in the security bureaucracy. The combination of central databases such as POLNET (Police Information Network) and Identity Notification System (KBS) in a common pool proves that the necessary institutional and technical ecosystem exists for person or place-oriented crime prediction algorithms to become fully operational in the near future in Turkey.
In criminal procedure law, the ability of law enforcement or judicial authorities to impose restrictive measures against an individual depends, as a rule, on the existence of the concept of reasonable suspicion based on concrete facts and evidence. Artificial intelligence-supported crime prediction systems somehow erode this concrete and action-oriented structure of the concept of suspicion. Algorithms produce algorithmic suspicion based on the correlation of past data, before any execution action has taken place in the concrete world.35
This situation damages the presumption of innocence, which is the most fundamental pillar of criminal law36. Profiling the individual as a potential suspect due to an act that has not yet been committed through artificial intelligence's statistical risk scoring, rather than an action that the individual carries out with his own free will, is almost a practice of reading intent. Criminal law is based on actual criminal law; The mentality of the perpetrator or the possible risk situation in the future cannot be a legitimate justification for criminal law sanctions or restrictive law enforcement measures. Making people objects of distrust in advance by reducing them only to risk scores directly harms the right not to be tarnished and human dignity.
This transformation created by algorithmic suspicion at the theoretical level, when evaluated in terms of Turkish criminal procedure law, creates a direct conflict with the 38 basic suspicion thresholds of the Criminal Procedure Code No. 5271 (CMK). CMK m. In order to initiate an investigation pursuant to Article 160/1, the law enforcement and the public prosecutor must reach an initial suspicion that gives the impression that a crime has been committed. Since the statistical risk scores or location estimates produced by the artificial intelligence algorithm are not based on a concrete basis, but are an inductive projection of past data, CMK m. It cannot constitute a concrete initial suspicion in the sense of 160.
Similarly, a serious legal barrier arises in the implementation of protection measures. For example, CMK art. According to Article 116, "reasonable suspicion" is required to conduct a search on a suspect's person, belongings or residence. A "black box" algorithm labeling a person as high risk through uncontrollable and non-individualized correlations, CMK art. It is far from meeting the reasonable suspicion criterion based on concrete facts sought by Article 116. Using algorithmic suspicion alone as a justification for a search, capture or physical pursuit by eroding the safeguards of criminal procedure is incompatible with the aim of reaching the material truth and protecting human dignity. For this reason, artificial intelligence outputs are not used as a degree of suspicion to replace the discretion of the judge or prosecutor in criminal procedure law; However, they should be considered as secondary intelligence clues that direct the law enforcement's concrete evidence collection activities.
The common perception that artificial intelligence systems produce completely objective, mathematical and unbiased decisions loses its validity in the face of empirical data. Algorithms are based on historical data sets (training data) loaded into them in machine learning processes. If this historical data contains biased practices of law enforcement against certain minorities or socio-economically disadvantaged groups, artificial intelligence learns and reproduces this discrimination as a mathematical reality. The clearest evidence of this structural flaw is that black defendants in the COMPAS system are erroneously found to be twice as risky.
This situation leads to a vicious circle called "Runaway Feedback Loop" or "Self-Fulfilling Prophecy" in the literature. The algorithm predicts neighborhood A as a “high-risk hotspot” due to historically biased data; allocates law enforcement resources to neighborhood A based on this estimate (over-policing); Increased police density in neighborhood A inevitably produces more arrests and records of minor crimes; These new records are loaded back into the system as data, and the algorithm announces how accurate its prediction is, legitimizing the bias under a technological mask.
Most of the predictive policing software is developed as a commercial product by private technology companies, and the operating logic, source codes and weight parameters of these algorithms are closed to judicial review on the grounds of trade secret (proprietary software).
A defendant who is arrested, whose bail request is rejected, or who is subjected to constant law enforcement surveillance due to an artificial intelligence risk score against him, does not have the opportunity to control what data was weighted and to what extent in the background of the system. Basing judicial decisions on algorithmic outputs whose rationale cannot be explained and whose source code cannot be inspected, eliminates the essence of the right to defense and reduces the state of law to a calculation bureaucracy.
In order for artificial intelligence-supported crime prediction systems to work, they depend not only on the existence of large data sets but also on the spread of these data in as many different areas as possible. The data collected in this context does not only consist of official criminal records and registry information. It also includes highly sensitive personal data that directly touches individuals' private lives, such as social media posts, security camera footage, financial transaction records, vehicle license plate recognition data and individuals' geographical location information41.
The uninterrupted data analysis capacity offered by artificial intelligence systematically erodes the boundaries of individual privacy by building a surveillance infrastructure on a social scale. This technological siege not only violates privacy; It creates a chilling effect that causes individuals to internalize the idea that they are constantly being watched42. As a result of this psychological pressure, individuals turn to behavioral self-censorship for fear of falling victim to an algorithmic misunderstanding or being included in the list of potential suspects, even if they are not involved in any crime43. The process in question; It causes the waiver of fundamental democratic rights such as freedom of expression, meetings and demonstrations, and ultimately the narrowing of the public space, which is an indispensable element of a democratic society. Therefore, predictive policing is not a mere security choice, but a new social regime tool in which the individual becomes an object of surveillance.
These structural dangers and fundamental rights violations caused by the uncontrolled use of artificial intelligence by law enforcement forces have necessitated strict legal regulations in the international arena. The Artificial Intelligence Regulation (EU) No. 2024/1689 (AI Act), which was adopted by the European Parliament and Council on 13 June 2024 and entered into force after being published in the EU Official Journal, is a landmark regulation in terms of legal history. The regulation in question subjected artificial intelligence technologies to a gradual system according to the risk levels they carry and determined the limits of the systems to be used in law enforcement activities with mandatory norms.
The AI Act prohibits law enforcement from using artificial intelligence systems that predict individuals' risk of committing a crime based solely on profiling or past criminal history, classifying them as "unacceptable risk" as a rule, due to their violation of human dignity and the presumption of innocence. This approach of the legislator clearly reveals that labeling the individual as a potential criminal based solely on statistical probabilities based on past data, without a concrete act or enforcement action reflected in the outside world, is incompatible with the principle of the rule of law. With this ban, it is aimed to prevent algorithmic systems from turning into a tool for reading prejudice or intent in criminal procedure.
In accordance with the Regulation, certain predictive policing infrastructures and bio-metric remote recognition systems, which are not completely indispensable in the fight against crime and are allowed exceptionally, are considered high-risk artificial intelligence systems. Extremely strict safeguards are provided for the use of these exceptionally permitted high-risk law enforcement systems. In this context, it has become a legal obligation for law enforcement units to conduct a comprehensive "Fundamental Rights Impact Assessment", which measures the risks that the system may pose to individuals' constitutional rights, before putting the system into operation. In addition, in order to prevent the transparency crisis and violations of the right to defense arising from the black box structure of the algorithms, technical explainability and logging standards have been regulated, making the operating logic of the system and data weighting criteria auditable.
The most vital legal barrier brought by the regulation is the principle of human supervision, which states that the outputs of the system can be inspected, filtered and overridden when necessary by a real human at every stage. In accordance with this principle, no risk score or analysis produced by artificial intelligence can constitute automatic grounds for enforcement or protective measures for law enforcement. The ultimate discretion and legal responsibility in procedural law cannot be delegated to algorithms; On the contrary, it should be subject to human control at every stage. Thus, the AI Act has built a global normative standard that protects human dignity and legal safeguards in the face of technological rationalization, preventing the automation of criminal law processes.
When the legal legislation in Turkey is examined, it is seen that there is a serious constitutional and legal gap regarding the control of artificial intelligence. Subparagraph (c) of the 1st paragraph of Article 28 titled "Exceptions" of the Personal Data Protection Law No. 6698 (KVKK); It has completely exempted from the scope of the law the processing of personal data within the scope of preventive, protective and intelligence activities carried out by public institutions and organizations authorized by law in order to ensure national defence, national security, public security, public order or economic security.
This unsupervised exemption regime in Türkiye clearly contradicts comparative law and especially the European Union acquis. While the European Union generally regulates the protection of personal data with GDPR (General Data Protection Regulation) 45, it has disciplined the data processing processes of law enforcement and judicial authorities in crime prevention, investigation and punishment activities with the Law Enforcement Directive (LED) No. 2016/680, which is completely customized for this field. The directive in question; It requires law enforcement units to comply with the principles of legality, transparency, data minimization and purpose limitation in data processing processes; It gives individuals the right to directly object to decisions based on automated data processing (such as profiling and crime prediction). Therefore, in EU law, law enforcement activities, far from being excluded from data protection safeguards, are subject to a much stricter and specific protection shield in order to prevent abuses. In Turkish law, KVKK art. The absolute exception brought by Article 28/1-ç causes a structural transparency crisis by completely depriving law enforcement forces' artificial intelligence-based crime prediction and big data analytics processes of constitutional data protection guarantees.
Removing law enforcement activities from the protective umbrella of KVKK with such a broad exception leaves artificial intelligence-based crime prediction and big data analytics processes completely uncontrolled and devoid of constitutional guarantees. In order to eliminate this structural danger, Türkiye urgently needs to enact an independent artificial intelligence law that is compatible with the EU AI Act norms and draws the limits of artificial intelligence use by law enforcement units47. With this legal regulation, person-oriented algorithmic profiling and precaution-taking processes should, as a rule, be subject to the approval of the criminal judge of peace, and the concept of suspicion should be prevented from becoming arbitrary through algorithmic rationalization.
The liability regime that will be implemented in case individuals are unjustly detained, arrested, or digitally marked and subject to rights restrictions as a result of artificial intelligence systems' false positive predictions, biased coding, or systemic malfunctions in the database should be clarified. In accordance with the last paragraph of Article 125 of the Constitution; "The administration is obliged to pay the damages arising from its own actions and transactions." .
Since artificial intelligence does not have an independent legal personality at the current technological level, law enforcement units that use these systems as a service tool in the execution of public services and therefore the Ministry of Internal Affairs will be directly responsible. Due to the autonomous character and black box structure of artificial intelligence, the obligation of the administration to prove the fault (service defect) cannot be placed on the aggrieved party. For this reason, the principles of strict liability and social risk in administrative law dogma should be applied in compensating damages arising from artificial intelligence-supported law enforcement activities; The individual's protection shield against technological risks should be expanded.
Predictive law enforcement activities; It provides mathematical predictions about the spatial and temporal focal points of crime by filtering enormous data masses through analytical filters and becomes an undeniable operational force multiplier in the protection of public order. However, these algorithmic projections created to fight crime carry the risk of structural erosion of the presumption of innocence and the principle of fault, which are the founding pillars of criminal law. Constructing the process on a narrative of potential guilt that is far from concrete foundations causes the state to remain on the text, eroding the constitutional guarantees it provides to the individual.
These digital interventions, which are carried out through algorithmic outputs without a concrete enforcement action in the sense of criminal procedure law or a criminal phenomenon reflected in the outside world, produce a modern practice of reading intentions in which the individual is transformed into an object of surveillance by deviating from the basis of actual criminal law. Mathematical algorithmic suspicion, which replaces the reasonable suspicion standard in criminal law, disconnects individuals from live events and reduces them to mere risk scores. This situation creates an irreparable crisis on the equality of arms and the right to a fair trial by profiling the criminal tendencies of the perpetrator in advance.
On the other hand, the black box structure inherent in predictive systems makes it difficult to audit the use of these technologies in administrative activities. Detection of structural errors and rights violations arising from the autonomous character of artificial intelligence-supported law enforcement activities is too complex to be resolved according to classical service defect principles. In this context, in cases of loss of rights arising from algorithmic illusions or data manipulations, placing the burden of proving fault on the aggrieved party will damage the sense of justice. Therefore, from the perspective of administrative law, the application of strict liability and social risk principles in compensating the risks posed by artificial intelligence-supported preventive law enforcement activities is an inevitable necessity in order to expand the legal protection shield of the individual against technological asymmetry.
Ultimately, big data and artificial intelligence integrations can be considered strategic instruments in preventing crime and establishing public safety; However, the limits of this use should be limited by the principles of transparency, explainability and absolute human supervision, as emphasized in the European Union Artificial Intelligence Act (AI Act) and human rights norms. Technological possibilities can only be a means of maintaining social peace and security; Individual safeguards, liberal freedoms and human dignity, which are the ultimate goals of criminal justice, should never be sacrificed to the statistical probability matrices of soulless algorithms.
"Problems in Practice and Legal Dimension", Journal of Liberal Thought, No. 1, 2025.
"Legal Liability", Dispute Court Journal, Vol. 7, No. 13, 2019.
Magazine, Vol. 9, No. 2, 2020.
Altun, Abdurrahim: Granting Electronic Personality to Artificial Intelligence, Ankara, Justice
Publishing House, 2023.
Academy Journal, Vol. 9, No. 35, 2018.
Bozkurt Yüksel, Armağan Ebru: "The European Commission's Artificial Intelligence Regulation
2022.
Justice Quarterly, Vol. 31, No. 4, 2014.
Publishing, 2023.
Faculty Journal, Vol. 28, No. 49, 2023.
https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=oj:JOC_2018_252_R_0026, 07.05.2026.
(Online) https://www.europarl.europa.eu/news/en/pressroom/20240308IPR19015/artificial- intelligence-act-meps-adopt-landmark-law, 08.05.2026.
(Online) https://www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-aiact-first- regulation-on-artificial-intelligence, 15.05.2026.
Gülel, Hazal: Artificial Intelligence from a Legal Perspective, Ankara, Adalet Publications, 2023.
Güner, Cemil: "Non-contractual claims arising from damage caused by artificial intelligence"
"Law Applicable to Liability", Journal of Dispute Court, Vol. 8, No.
15, 2020.
Management Review, Vol. 61, No. 4, 2019.
Hakeri, Hakan: General Provisions of Criminal Law, 25th ed., Ankara, Adalet Publishing House,
2022.
Part, Istanbul, Beta Publishing, 2004.
Karabal, Ahu: "Law Violations in Artificial Intelligence", Artificial Intelligence and Philosophy of Mind
Magazine, Vol. 6, No. 2, 2023.
"A Review in the Light of the Directive", Ankara Journal of European Studies, Vol. 22, No.
1, 2023.
Anadolu Strategy Magazine, Vol. 2, No. 1, 2020.
Kayıhan, Şaban: Personal Law, Ankara, Seçkin Publishing, 2022.
Publishing, 2023.
Ankara, Seçkin Publishing, 2023.
Kulular İbrahim, Merve Ayşegül: Planned Obsolescence from a Legal Perspective, Ankara, Justice
Publishing House, 2021.
Oğuzman, M. Kemal et al.: Personal Law, Istanbul, Filiz Kitabevi, 2024.
Publishing, 2024.
Berkeley Technology Law Journal, Vol. 31, No. 1, 2017.
Popp, Tony: "Black Box Justice", The Pennsylvania Gazette, Vol. 115, No. 3, 2017.
-p1ai/news/item/19639138/, 15.05.2026.
Threats", Medeniyet Cultural Research Belleteni, Vol. 4, No. 7, 2025.
and Regulation, 2022.
Law", Cardozo Law Review, 2018.
Publishing, 2018.
1 Hazal Gülel, Artificial Intelligence from a Legal Perspective, Ankara, Adalet Publications, 2023, p. 25.
2 Ahu Karabal, "Law Violations in Artificial Intelligence", Journal of Artificial Intelligence and Philosophy of Mind, Vol. 6, No. 2, December 2023, p. 122. 3 Cemil Güner, "Law Applicable to Non-Contractual Liability Arising from Damage Caused by Artificial Intelligence", Journal of Dispute Court, Vol. 8, No. 15, June 2020, p. 231.
4 EUROPARL, "Artificial Intelligence Act: MEPs adopt landmark law", (Access Date: 08.05.2026). (Online), https://www.europarl.europa.eu/news/en/pressroom/20240308IPR19015/artificial- intelligence-act-meps-adopt-landmark-law. 5 Gift Ebru Bozkurt Yüksel, "An Overview of the European Commission's Artificial Intelligence Regulation Proposal", Turkish Justice Academy Journal, Vol. 13, No. 51, July 2022, p. 21.
6 EUROPARL, "Artificial Intelligence Act: MEPs adopt landmark law",(E.T.: 08.05.2026), (Online)https://www.europarl.europa.eu/news/en/pressroom/20240308IPR19015/artificial-intelligence-act-meps-adopt-landmark-law.
7 Süleyman Dost, "Artificial Intelligence and Freedom of Expression", Dicle University Faculty of Law Journal (DÜHFD), Vol. 28, P. 49, December 2023, p. 284.9 10
8 Shlomit Yanisky Ravid, Xiaoqiong (Jackie) Liu, “When Artificial Intelligence Systems Produce Inventions: The 3A Era and An Alternative Model For Patent Law,” Cardozo Law Review , 2018, p. 2223.
9 Michael Haeinlein, Andreas Kaplan, “A Brief History of Artificial Intelligence: On the Past, Present, and Future of Artificial Intelligence,” California Management Review , Vol. 61, P. 4, July 2019, p. 6.
10 Sinan Sami Akkurt, "Legal Liability Arising from the Autonomous Behavior of Artificial Intelligence", Journal of Dispute Court, Vol. 7, No. 13, June 2019, p. 41. 11 Akkurt, "Autonomous Artificial Intelligence...", p. 43; Haeinlein, Kaplan, "A Brief History...", p. 39. 12 13 12 Başak Bak, "Legal Status of Artificial Intelligence in Terms of Civil Law and Legal Liability Arising from the Use of Artificial Intelligence", Turkish Justice Academy Journal, Vol. 9, P. 35, July 2018, p. 213-215.
13 EUROPARL, "EU AI Act: First Regulation on Artificial Intelligence",(E.T.: 15.05.2026), (Online)https://www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-aiact-first- regulation-on-artificial-intelligence.
14 Akkurt, "Autonomous Artificial Intelligence...", p. 44.
15 OG, 08/12/2001 – 24607, Turkish Civil Code No. 4721.
16 Şaban Kayıhan, Personal Law, Ankara, Seçkin Publishing, 2022, p. 19. 17 M. Kemal Oğuzman et al., Personal Law, Istanbul, Filiz Kitabevi, 2024, p. 213. 19 21 23
18 See, "From the Perspective of Civil Law...", p. 219.
19 European Parliament resolution of 16 February 2017 with recommendations to the Commission on Civil Law Rules on Robotics (2015/2103(INL))”, (E.T.:7.05.2026.), (Online), https://eur- lex.europa.eu/legal-content/EN/TXT/?uri=oj:JOC_2018_252_R_0026. 20 Abdurrahim Altun, Granting Electronic Personality to Artificial Intelligence, Ankara, Adalet Publishing House, 2023, p. 198.
21 A. Dilşad Keskin, Huriye Reyhan Demircioğlu, Civil Law II, Ankara, Seçkin Yayıncılık, 2023, p. 30.
22 Merve Ayşegül Kulular İbrahim, Planned Obsolescence in Legal Perspective, Ankara, Adalet Publishing House, 2021, p. 287. 23 See, "From the Perspective of Civil Law...", p. 217-218. 25
24 İzzet Özgenç, Turkish Criminal Law General Provisions, 19th Edition, Ankara, Seçkin Yayıncılık, 2024, p. 45. 25 Mahmut Koca, İlhan Üzülmez, Turkish Criminal Law General Provisions, 16th Edition, Ankara, Seçkin Yayıncılık, 2023, p. 32-33.
26 Hakan Hakeri, Criminal Law General Provisions, 25th Edition, Ankara, Adalet Publishing House, 2022, p. 58.
27 Kayıhan İçel, Süheyl Donay, Comparative and Applied Criminal Law General Part, Istanbul, Beta Publishing, 2004, p. 74.
28 Caner Yenidunya, Risk Society and Criminal Law, Istanbul, Oniki Levha Publishing, 2018, p. 112 ff. 29
29 Nurettin Alkan, Yunus Emre Karamanoğlu, "Preventive Law Enforcement Based on Predictive Law Enforcement: Examples from the Russian Federation", Journal of Security Sciences, Vol. 9, No. 2, 2020, p. 387-418. 30 31
30 Zuzanna Warso, “Human Rights Requirements for Person-Based Predictive Policing: Lessons from Selected ECtHR Case Law and its Limits,” Technology and Regulation, 2022, p. 71-73.
31 RF MINISTRY OF INTERNAL AFFAIRS, "V Moskvesistema raspoznavaniya lits za sutki vyyavila bolee 200 narushiteley rezhima samoizolyatsii",(E.T.: 15.05.2026), (Online) https://xn--b1aew.xn-- p1ai/news/item/19639138/; Nurettin Alkan, Yunus Emre Karamanoğlu, "Preventive Law Enforcement Based on Predictive Law Enforcement 32 33 34: Examples from the Russian Federation", Journal of Security Sciences, Vol. 9, No. 2, 2020, p. 387-418.(E.T.:17.05.2026) 32 Ozan Kavsıracı, Mehmet Demirbaş, “Examination of Intelligence Activities in Terms of State Security”, Anadolu Strategy Magazine, Vol. 2, No. 1, 2020, p. 49.
33 Anthony A. Braga, Andrew V. Papachristos, David M. Hureau, “The Effects of Hot Spots Policing on Crime: An Updated Systematic Review and Meta-Analysis,” Justice Quarterly , Vol. 31, No. 4, 2014, p. 633.
34 Niyazi Umut Akıncıoğlu, “Artificial Intelligence Supported Crime Prediction Systems: Problems in Practice and Legal Dimension”, Liberal Thought Magazine, No. 1, December 2025; Buse Nur Tufan, “Artificial Intelligence and Crime: 37 Legal and Ethical Threats for the Future”, Medeniyet Cultural Research Belleteni, Vol. 4, P. 7, 2025. 35 İzzet Özgenç, Turkish Criminal Law General Provisions, 19th Edition, Ankara, Seçkin Yayıncılık, 2024, p. 45.
36 Nur Centel, Hamide Zafer, Criminal Procedure Law, 20th Edition, Istanbul, Beta Publishing, 2023, p. 450 ff.
37 Caner Yenidunya, Risk Society and Criminal Law, Istanbul, Oniki Levha Yayıncılık, 2018, p. 112 ff. 38 OG, 17/12/2004 – 25673, Criminal Procedure Law No. 5271 39 40
39 Tony Popp, “Black Box Justice,” The Pennsylvania Gazette , Vol. 115, No. 3, 2017, p. 38-47. 40 Hazal Gülel, Artificial Intelligence from a Legal Perspective, Ankara, Adalet Publications, 2023, p. 25. 41 Sarah Brayne, “Big Data Surveillance: The Case of Policing,” American Sociological Review , Vol. 82, No. 5, 2017, p. 977-1008.
42 ECHR, Big Brother Watch and Others v. United Kingdom [GC], App. No: 58170/13, 62322/14 and 24960/15, 25.05.2021, (E.T.: 16.06.2026.), (Online) https://hudoc.echr.coe.int/eng?i=002-12080, para.10.
43 Jonathon W. Penney, “Chilling Effects: Online Surveillance and Wikipedia Use,” Berkeley Technology Law Journal , Vol. 31, No. 1, 2017, p. 117 ff. 44 46
44 Enver Kaşlı, “Protection of Personal Data in Law Enforcement Activities: A Review in the Light of the European Union Europol Regulation No. 2016/794 and Law Enforcement-Criminal Justice Directive No. 2016/680”, Ankara Journal of European Studies, Vol. 22, No. 1, 2023, p. 97-115.
45 General Data Protection Regulation (GDPR). (n.d.). General Data Protection Regulation (GDPR), (Online) https://gdpr-info.eu/ (E.T.: 17 June 2026).
46 European Parliament and Council, “Directive (EU) 2016/680 of the European Parliament and of the Council of 27 April 2016”, (E.T.: 17.06.2026.), (Online) https://eur- lex.europa.eu/eli/dir/2016/680/oj/eng. 48
47 Abdurrahim Altun, Granting Electronic Personality to Artificial Intelligence, Ankara, Adalet Publishing House, 2023, p. 198.
48 Cemil Güner, “Law Applicable to Non-Contractual Liability Arising from Damage Caused by Artificial Intelligence”, Journal of Dispute Court, Vol. 8, No. 15, June 2020, p. 231.
KÜHEYLAN HUKUK