Automated Law Enforcement and Human Rights: Should We Eliminate Crime Through Technology?
- Esmée Cadogan
- Jun 26
- 8 min read
Introduction
Automated law enforcement of the criminal law promises to shift enforcement from reacting to wrongdoing after the event to preventing it altogether. This blog argues that, although such systems may seem attractive where technology can reduce harm, they raise serious concerns when implemented through impossibility structures. In particular, automated enforcement may intensify the effects of over-inclusive offences (offences that criminalise more conduct than is necessary to target the relevant harm) by removing proportionality assessments and prosecutorial discretion. Automated enforced may also result in prior restraint, especially where speech or protest is blocked before it can occur. The broader concern is that these systems may alter the relationship between law, rights and democratic contestation in ways that are difficult to justify.
Scope of Automated Law Enforcement
Currently, laws are predominantly enforced after the crime has been committed. This means it is possible to choose to break the law and risk facing the relevant sanction. However, automated law enforcement goes beyond automated detection and aims to prevent unlawful conduct before it happens. This typically relies on impossibility structures, which are defined as systems that impose technological constraints on the ability to perform prohibited actions.[i] An example of an impossibility structure being developed is the Driver Alcohol Detection System for Safety, which aims to prevent vehicles from moving if the driver exceeds the blood alcohol limit.[ii]
Technological limits once meant that automated law enforcement seemed possible only for narrow, easily measurable rules. However, advances in artificial intelligence may soon allow much broader forms of automated enforcement, including open-textured laws stated in general and evaluative terms (such as ‘reasonable’ or ‘offensive’).[iii] In time, systems powered by more capable AI could block or interrupt digital activities before the prohibited act is carried out, rather than merely detecting them after the event.[iv] Furthermore, as smart devices become embedded in everyday objects, this kind of control could reach far beyond the online world, significantly reducing the practical space for criminal conduct.’[v]
Overinclusive Offences
Since automated law enforcement would apply automatically to any conduct identified as violating the law, it could undermine safeguards that mitigate the effect of over-inclusive laws. For example, DPP v Ziegler held that some offences require a case-specific assessment of whether the interference with human rights is necessary in a democratic society.[vi] An example of this is the offence of sending messages that are ‘indecent or grossly offensive’ under the Malicious Communications Act 1988, which has been held to require a case-specific proportionality assessment when the defendant’s Article 10 rights are engaged.[vii] This proportionality safeguard is rendered ineffective if the individual is preemptively prevented from making the communication. A possible solution is to ensure all criminal offences are narrow and precise enough that proof of the ingredients of the offence will, without more, be sufficient to render a conviction proportionate, as was the case in SAZ.[viii]
This highlights a broader point that a change in our enforcement methods may necessitate a change in the criminal law’s scope. Currently, overinclusive offences do not necessarily lead to disproportionate interferences with human rights and liberties. This is because limited police and prosecutorial capacity means that existing enforcement mechanisms inevitably become selective in deciding which cases are worth pursuing.[ix] However, if such laws were enforced through impossibility structures, these same practical constraints do not apply.[x] Therefore, any implementation of automated law enforcement would require heightened scrutiny of the scope of criminal offences to compensate for the elimination of downstream safeguards like proportionality assessments and prosecutorial discretion.
Prior Restraint
Another human rights concept that could be developed to address this issue is the prohibition of prior restraint. Prior restraint is where state authorities from checking and censoring speech prior to its publication.[xi] On the one hand, in a case reviewing whether the Directive on Copyright in the Digital Single Market—which incentivises online platforms to use filters preventing the upload of copyright content—was compatible with the right to freedom of expression, the European Court of Justice acknowledged that prior restraint ‘pose[s] such a risk to compliance with that fundamental right that a particularly tight legal framework is required’.[xii] On the other hand, scholars like Bendor argue that prior restraint is more necessary in the digital age because the Internet’s ‘ease and immediacy of publication’, its ‘eternal exposure’, and its capacity for ‘global virality’ greatly magnify the harms of unprotected speech and render ex post criminal sanctions inadequate.[xiii]
Although Bendor’s argument has force where new technology makes unlawful activity easier, necessitating technological tools to combat it,[xiv] it is not as persuasive when applied to impossibility structures, which aim to eliminate rather than reduce the relevant criminal activity and thus are likely to have far greater adverse consequences. Any legal regime will inevitably contain imperfections, and perfect enforcement of these imperfect laws can stifle the usual mechanisms through which the law is challenged and evolves. Llansó argues that automated forms of prior restraint, such as filtering and suppressing speech before it enters the public sphere, do not merely risk over-removal but also erode the conditions for democratic critique, because opaque algorithmic enforcement prevents both speakers and the public from knowing what expression is being excluded and from reassessing whether existing legal boundaries of permissible speech remain justified in that context.[xv]
A well-functioning democracy depends, at least to some extent, on the possibility of civil disobedience, which Arendt argues is a necessary means by which citizens can signal deep structural injustices when ordinary legal channels fail.[xvi] By annihilating the very acts through which citizens can contest and expose unjust or outdated laws, impossibility structures risk foreclosing democratic deliberation and the law’s capacity for self-correction. Therefore, it may be more proportionate to prohibit the use of impossibility structures for the prior restraint of certain types of conduct, such as protests that engage fundamental rights, even if such conduct is unlawful. This is because sanctioning such conduct afterwards allows the state to protect its legitimate interests (e.g., in public order) whilst still allowing conduct challenging the existing laws to occur, facilitating democratic dialogue about whether the current legal position is correct.
Conclusion: The Fundamental Problem with Automated Law Enforcement
In summary, this blog has argued that automated law enforcement creates two interconnected risks. It may lead to overenforcement by applying overinclusive offences without the moderating influence of prosecutorial discretion or proportionality review, and it may function as a prior restraint by preventing conduct, particularly manifestations of freedom of expression, before it can be evaluated within legal processes. Both concerns arise from the use of impossibility structures that eliminate ex post safeguards traditionally embedded in the criminal justice system.
Overall, these concerns about overenforcement and prior restraint point towards a more fundamental problem: if law transitions from a system primarily based on voluntary compliance to one based on compulsion through impossibility structures, the need to have public support for the criminal law diminishes. Whilst legal compliance is typically desirable, legal and institutional mechanisms may sometimes fail to secure fundamental rights. In such circumstances, the recognised possibility of resistance functions as an ultimate safeguard, preserving the normative force of human rights by ensuring that persistent injustice is not rendered immune from challenge.[xvii] If such unlawful resistance efforts could be successfully preempted through automated law enforcement, then it would be far easier for an authoritarian regime to impose laws on an unwilling population.[xviii] Therefore, the establishment of automated law enforcement infrastructure may be concerning even if it is initially used to enforce reasonable laws. This is because, by removing the possibility of resistance, the overriding safeguards against an authoritarian drift are reduced.
Author bio
Esmée Cadogan is a first year undergraduate, studying Law at Trinity College (University of Cambridge) whose interest in the field was sparked by a research fellowship with Leaf on AI Safety and Governance, as well as their general enthusiasm for human rights law.
References
[i] Antje von Ungern-Sternberg, ‘Automated law enforcement: perfect vision or dystopia?’ in Research Handbook on the Law of Artificial Intelligence, ed. by Woodrow Barfield and Ugo Pagallo (Cheltenham: Edward Elgar, 2025), pp. 250–274 (p. 250). Impossibility structures can also work for omissions by fulfilling a legal obligation automatically, and thereby removing the choice not to comply with them. One example of this is the development of smart contracts, which use blockchain technology to carry out agreed actions automatically once specified conditions are met. This means that outcomes such as transferring money or granting access happen automatically without requiring human approval or intervention: Andres Guadamuz, “All Watched over by Machines of Loving Grace: A Critical Look at Smart Contracts,” Computer Law & Security Review, 35.6 (2019), p. 105338, doi:10.1016/j.clsr.2019.105338.
[ii] Susan A. Ferguson, Abdullatif (bud) Zaouk, and Clair Strohl, “Driver Alcohol Detection System for Safety (DADSS). Background and Rationale for Technology Approaches,” SAE Technical Paper Series (400 Commonwealth Drive, Warrendale, PA, United States).
[iii] For developments in AI legal reason capabilities and current limitations see, for example: Eljas Linna and Tuula Linna, “Judicial Requirements for Generative AI in Legal Reasoning,” in arXiv [Cs.AI], 2025 <http://arxiv.org/abs/2508.18880> accessed 29 December 2025; Albert Sadowski and Jaroslaw A. Chudziak, “On Verifiable Legal Reasoning: A Multi-Agent Framework with Formalized Knowledge Representations,” Proceedings of the 34th ACM International Conference on Information and Knowledge Management (New York, NY, USA), 2025, pp. 2535–45.
[iv] An example of technological advance that could achieve this would be the development of Artificial General Intelligence which is AI that can exhibit or simulate behaviour as intelligent as, or more intelligent than, that of a human being. For further information about when and how Artificial General Intelligence might be developed see: Gil Syswerda, “Timeline to Artificial General Intelligence 2025 – 2030+,” SuperIntelligence - Robotics - Safety & Alignment, 2.3 (2025), doi:10.70777/si.v2i3.15119.
[v] The incorporation of smart technology into physical objects is known as the Internet of Things: Himanshu Singhvi, “The Internet of Things in 2025: Trends, Business Models, and Future Directions for a Connected World,” INTERNATIONAL JOURNAL OF INTERNET OF THINGS, 3.1 (2025), pp.17–24, doi:10.34218/ijiot_03_01_003.
[vi] Director of Public Prosecutions v Ziegler [2021] UKSC 23, para.70.
[vii] R v Casserly [2024] EWCA Crim 25, para.48. See also: Connolly v Director of Public Prosecutions [2007] EWHC 237 (Admin).
[viii] Re Abortion Services (Safe Access Zones) (Northern Ireland) Bill [2022] UKSC 32, para.34.
[ix] Woodrow Hartzog, “Inefficiently Automated Law Enforcement,” Michigan Law Review, 120.8 (2022); Erik Luna, “Prosecutorial Decriminalization,” Journal of Criminal Law & Criminology, 2012.
[x] Antje von Ungern-Sternberg, ‘Automated law enforcement: perfect vision or dystopia?’ in Research Handbook on the Law of Artificial Intelligence, ed. by Woodrow Barfield and Ugo Pagallo (Cheltenham: Edward Elgar, 2025), pp. 250–274 (p. 255).
[xi] This doctrine is part of, inter alia, German and American constitutional laws. Thomas I. Emerson, “The Doctrine of Prior Restraint,” Law and Contemporary Problems, 20.4 (1955), p. 648, doi:10.2307/1190292.
[xii] ECJ, Judgment of 26 April 2022, Poland/EP and Council, C-401/19, para.68.
[xiii] Ariel L. Bendor and Michal Tamir, “Prior Restraint in the Digital Age”, 27 Wm. & Mary Bill Rts. J. 1155 (2019).
[xiv] “Research on the Copyright Content Filtering Obligation of Internet Service Providers,” International Journal of Frontiers in Sociology, 6.1 (2024), doi:10.25236/ijfs.2024.060104.
[xv] Emma J. Llansó, “No Amount of ‘AI’ in Content Moderation Will Solve Filtering’s Prior-Restraint Problem,” Big Data & Society, 7.1 (2020), doi:10.1177/2053951720920686.
[xvi] Hannah Arendt, Henry David Thoreau, and Roger Berkowitz, On Civil Disobedience (Library of America, 2024). See also: Ten-Herng Lai, “Civil Disobedience, Costly Signals, and Leveraging Injustice,” Ergo (Ann Arbor, Mich.), 7.0 (2021), doi:10.3998/ergo.1137.
[xvii] Gwilym David Blunt, ‘The Right to Resistance’, in Global Poverty, Injustice, and Resistance, ed. by Gwilym David Blunt (Cambridge: Cambridge University Press, 2019), pp. 41 – 70.
[xviii] Xu Xu, “To Repress or to Co‐opt? Authoritarian Control in the Age of Digital Surveillance: Authoritarian Control in the Age of Digital Surveillance,” American Journal of Political Science, 65.2 (2021), pp. 309–25, doi:10.1111/ajps.12514; Matthew J. Tokson, “Artificial Intelligence and the Anti-Authoritarian Fourth Amendment,” 2025, doi:10.2139/ssrn.5182213; H. Akin Ünver, Artificial intelligence (AI) and human rights: Using AI as a weapon of repression and its impact on human rights, In-Depth Analysis (Publications Office of the European Union, 2024) <https://www.europarl.europa.eu/thinktank/en/document/EXPO_IDA(2024)754450> accessed 29 December 2025.
Comments