The European PNR Directive as an instance of pre-emptive, risk-based algorithmic security and its implications for the regulatory framework¹

2.1International bilateral PNR agreements involving the European Union (EU) and global developments

The focus of this article is on the European PNR framework as it is established by Directive (EU) 2016/681. The previously existing extra-European legislation and PNR international agreements are thus not the specific focus of this analysis. However, a short overview of the international and global landscape of PNR regulation will help clarify and contextualise the EU case and especially the role of the EU as, at once, a ‘norm-taker’ (Argomaniz, 2009) and an active contributor to the current process leading to the emergence of a global PNR regime.

In the direct aftermath of 9/11, the United States (US) adopted the Aviation and Transportation Security Act, which obliged air carriers to provide US customs, border control and security authorities access to the passengers’ data contained in their booking and departure control systems for flights to and from the US (Wojnowska-Radzińska, 2021, p. 117). As an upshot of this US legislation, a transatlantic agreement with the EU was needed in order to regulate the transfer of data by European Air carriers. A first US-EU Agreement was signed in 2004, but was shortly after declared void by the Court of Justice of the European Union (CJEU) for relying on the wrong legal basis. A new version of the agreement was signed in 2007 between the EU and the US, but data protection concerns lead to the drafting of a new agreement that was finalised in 2011 and signed by the EU Parliament in 2012. This agreement, still in force, requires and regulates the transfer of PNR data from the EU to the US, but not vice versa.

In addition to the EU-US PNR treaty, the EU negotiated bilateral PNR agreements with Australia (signed in 2011) and Canada (signed in 2006 and re-negotiated in 2014). The agreement with Canada, however, has never entered into force. Indeed, based on an EU-Parliament request issued in 2014, the CJEU decided in 2017 that some of the agreement’s norms conflicted with the EU’s Charter of Fundamental Rights (CFR; CJEU, 2017; Carpanelli & Lazzerini, 2017; Tambou, 2018). Negotiations for additional PNR agreements were launched in 2015 and 2020 with Mexico and Japan respectively (EU Commission, 2015; Maruhashi, 2020).

The existence of international PNR agreements, particularly the EU-US agreement, was a driving motivation for creating a European PNR regulation (Argomaniz, 2009; Bigo et al., 2015; Boehm, 2011). Indeed, EU authorities themselves did not have automatically access to the PNR data transferred to the US (De Hert & Papakonstantinou, 2009, p. 369; Blasi Casagran, 2015). Once the EU PNR Directive regulation was in force, it in turn triggered reciprocity claims by third states and led to requests for further bilateral PNR agreements (EU Commission, 2020b, p. 49).

Recently, the use of PNR data has been at the core of regulations at the global level as well. In 2017 and 2019, the Security Council (SC) of the United Nations has passed two resolutions which oblige states to collect, retain and transfer PNR data.22 Originally, the duty was limited to the purposes of combating terrorism, but it has subsequently been expanded to include organised crime. Following the SC Resolutions, the International Civil Aviation Organisation adopted international standards in 2020 for the use of PNR data with binding character (EU Commission, 2020b, pp. 6–7).

2.2The EU PNR Directive

The first proposal to introduce a PNR regulatory framework in the EU dates back to 2007, with a proposal by the EU Commission for a PNR Council decision (EU Commission, 2007). The Commission’s proposal, however, lost its validity, due to modifications in the legislative procedures introduced by 2009’s Lisbon Treaty. In the same year, the Council urged the Commission to prepare a new proposal in accordance with the new legal framework (European Council, 2009). This was presented in 2011 and contained provisions modelled on the EU-US PNR agreement (EU Commission, 2011). The proposal met with criticism from the EU Parliament, who rejected it in 2013 and thus temporarily halted the regulatory plans for introducing a European PNR framework (EU Parliament, 2013).

Following the terrorist attacks in Paris in January 2015, however, the Commission’s proposal was presented anew. In the mutated political climate, marked by a widespread sense of insecurity, the EU Parliament was more inclined to accept security measures restricting fundamental liberties than two years before (Bigo et al., 2015) and, in 2016, the EU PNR Directive was finally adopted.

The PNR Directive entered into force in May 2016. At present, all EU Member States except Denmark have implemented the Directive. Denmark, as a consequence of its opting-out of the measures in the ‘Area of Freedom, Security and Justice’, is not bound by the Directive, while the United Kingdom (UK) was bound by the Directive until 31 January 2020.33 The UK, however, joined the EU PNR regime on the basis of a separate agreement and Denmark put in place national legislation establishing an EU-like PNR regime (CJEU Avocate Général, 2022, fn 60).

The directive introduces a pre-emptive, risk-based approach to security that is a novelty in the landscape of EU data exchange. Indeed, according to its own text, the specific innovation of the intended use of PNR data consists in enabling member states to identify individuals who are potentially involved in serious crimes or terrorist activities but whose involvement was not previously known or suspected (PNR Directive, Nr. 6 and 7).

To that end, EU member states must oblige air carriers operating extra-EU flights to transmit the data collected as part of the reservation and check-in process to a central national Passenger Information Unit (PIU) run by the LEAs (usually the national or federal police) of the EU state from which the flight departs or that of its destination. These data are analysed by the PIU and can be transmitted to the central units of other EU member states, to Europol and to third countries as well. According to the PNR Directive, the processing of these data is limited to the purposes of ‘preventing, detecting, investigating and prosecuting terrorist offences and serious crime’ (Art. 1). The collection and processing of PNR data is mandatory only for extra-EU flights, but member states have the option of extending it to EU-internal flights (Art. 2). All but one of the member states that have implemented the PNR Directive so far have taken this option (EU Commission, 2020a, p. 10).

The passengers’ data transmitted by air carriers to the PIUs consist mostly of data that the carriers collect as part of their usual booking and check-in procedures. These data include name and contact information of the passenger, date of reservation and of planned travel, itinerary, travel agency, payment information, seat number and other seat information, complete baggage information, potential other persons travelling jointly with the passenger and a free-text field for entering ‘general information’ (Annex I of the PNR Directive).

National PIUs check the received information against European and international criminal databases, such as the Schengen Information System (SIS). Additionally, the PIUs analyse the data received based on risk criteria established by the PIU itself in order to assess the risk-level of each passenger. The risk criteria, which are treated as classified, may at any time be adapted in response to the PNR data processed (Art. 6, Nr. 2, lit. c). Passengers whose scores qualify them as high-risk, and who still qualify as such after a manual check of their data processing results, are selected for further examination (Art. 6).

2.3First implementation report and pending ruling by the CJEU

Two years after the expiration of the deadline for member states to transpose the PNR Directive into national law, the EC published a review report in accord with the provisions of Article 19 of the PNR Directive (EU Commission, 2020a). For the purposes of this review report, member states were to provide the EC with yearly statistics (Art. 20 PNR Directive).

The Commission’s evaluation of the first two years is positive. According to the report, the statistics provided by member states indicate that PNR data are a useful and effective means for investigating and prosecuting serious crimes and terrorism. However, the Commission itself relativises this assessment shortly after:

Accordingly, the report does not provide critical quantitative data such as the total number of datasets transferred to the national units, the number of passengers that have undergone additional screenings, the number of successful arrests or the share of false positives and false negatives. An independent assessment of the effectiveness, necessity and proportionality of the use of PNR data so far is therefore not possible.

The necessity and proportionality of the PNR measures have been questioned in national court proceedings that have led to requests for preliminary rulings by the CJEU, submitted in 2019 by the Belgian Constitutional Court and in 2020 by a German local court. Most of the questions posed to the CJEU concern the compatibility of the uses of PNR data with Articles 7 (Respect for private and family life) and 8 (Protection of personal data) of the CFR (Belgian Constitutional Court, 2019; see also Roda, 2020).

On January 27th 2022, the Advocate General of the CJEU released his preliminary opinion on the case submitted by the Belgian Constitutional Court. According to it, specific provisions of the Directive, for instance the ones allowing for a generalised retention of non-anonymised data or the collection of ‘general information’, must be modified or are to be declared invalid. However, the collection and processing of passengers’ data envisaged by the Directives are not as such considered to be incompatible with Art. 7 and 8 of the CFR. This applies, according to the Advocate General, also to the check against pre-defined criteria for classification into risk categories (CJEU, 2022; Thönnes, 2022).

3.1The PNR Directive as an instance of ‘algorithmic security’

In a recent article, Karen Yeung defined ‘algorithmic regulation’ as referring to

Yeung’s conceptualisation has been discussed with reference to different application areas (Eyert et al., 2022; Festic, 2022; Busch & De Franceschi, 2021, Peeters & Schuilenburg, 2018), yet the exploration of the concept for ‘critically interrogating’ the use of algorithms in the security domain is still quite novel. Interestingly, existing literature explicitly connecting the debate on algorithmic regulation with security research often focuses on PNR measures. Lena Ulbricht (2018) used Yeung’s conceptual framing to analyse the German parliamentary debates which led to German law’s implementation of the PNR directive. Similarly, it is through the lenses of Yeung’s conceptual framework that Bellanova and de Goede (2022) reflected on the transatlantic regulations between the EU and the US on the exchange of PNR data and the Terrorism Financing Tracking Program. The present article aims to continue this research on ‘algorithmic security’ (Bellanova and de Goede, 2022, p. 102) by focussing on the specific European PNR framework as it was set up by the PNR Directive. It thus situates itself on a middle ground between Ulbricht’s focus on national regulation, and Bellanova and de Goede’s transatlantic scope.

The European PNR Directive, like the German and transatlantic PNR regulations, can be fruitfully understood as an instance of algorithmic regulation. Indeed, it aims to provide authorities support as they decide which individuals pose a sufficiently high risk to warrant closer analysis. The system’s suggestions are based on ‘knowledge’ generated computationally through the analysis of the PNR data and aim to reach a broadly (and quite vaguely) pre-defined goal consisting in fighting terrorism and criminality. In the following, I will set out to explore the distinctive characteristics and implications of the European PNR mode of algorithmic regulation.

3.2Pre-emption and behavioural profiling

According to Yeung, algorithmic regulation systems can be reactive or pre-emptive, depending on the use they make of existing data. While reactive systems are based on the analysis of historical data, pre-emptive systems assess historic data in order to infer predictions about future behaviour (Yeung, 2018, p. 509). The EU PNR system incorporates both reactive and pre-emptive elements. Indeed, it uses historic data to detect existing violations or relevant past behaviour (by checking against existing official databases), but also to predict future behaviour. The main innovation of the PNR Directive derives from this second characteristic and consists in the introduction of a pre-emptive, risk-based approach to intra-European information exchange for security purposes. In addition to being checked for compliance with pre-established and known rules (not travelling with fake documents, not having committed a crime in the past, etc.), passengers are now screened in order to obtain a probabilistic assessment of their future behaviour. This approach is based on the assumption that mundane travel-related information can provide indications of a traveller planning or being involved in criminal or terrorist activity.

Yeung (2018) has further suggested that algorithmic regulation can be understood as a form of ‘surveillant-driven social sorting’ (512 f.). This general characterisation fits the EU PNR framework. Bellanova and Duez (2012, p. 122) have noted, moreover, that the PNR setting entertains a ‘symbiotic’ relationship with commercial socio-technical assemblages. These observations can be further substantiated by analysing the distinctive way in which the EU PNR framework introduces economically-oriented social sorting practices into the security domain.

Although personal data have been employed for ICT-supported categorisations since the last decade of the 20th century at least (Gandy, 1993; Lyon, 2003), a critical change occurred at the turning of the millennium. Shoshana Zuboff (2019, 2015) provides an insightful description of this turn, which has gone hand in hand with the rise of a new economic system she calls ‘surveillance capitalism’. This new economic logic relies on the appropriation of the raw, collateral data produced by internet users and their transformation into behavioural predictions (Zuboff, 2019, p. 74). The basic logic of the pre-emptive approach of the PNR Directive is symmetric to the basic functioning of the surveillance economy as described by Zuboff. In the same way as private companies appropriate the information surplus created by non-commercial activities for commercial purposes, the PNR system is based on the appropriation of the information surplus produced by law-abiding and fully legal activities (such as booking a flight ticket) for security purposes.

Take, for instance, a passenger who buys a flight ticket shortly in advance of a flight, pays it in cash instead of by credit card and checks luggage weighing 30 kg for a 3-day roundtrip. Such bits of information are provided by travellers in order to fly and processed by airlines in order to provide their services as air carriers. Within the PNR framework, they are appropriated by public authorities and used for (secondary) security purposes. In this new context, they are assumed to provide assessments of potential criminal or terrorist activity or intent. Although the bits of information mentioned above depict fully legal behaviour, they can, in combination, lead to a person be selected for further scrutiny. Indeed, they can be interpreted as indicating unusual, illogic or ‘abnormal’ behaviour. Additionally, they may be associated with behavioural patterns attributed to persons involved in criminal or terrorist activities, who, for instance, purchase a roundtrip but do not intend to fly back and take therefore much more luggage than they would need for a short trip.44

3.3Risk-based security and its paradoxical normativity

The pre-emptive approach embodied by the EU PNR framework is tightly intertwined with a risk-based approach. The affinity between risk-based and algorithmic regulation has been highlighted by Yeung (2018, p. 511), as well. On her view, algorithmic regulation can be seen as a form of prioritisation based on classification into risk-levels and aimed at providing decision-making support in the allocation of enforcement resources. As has been noted, risk-based rationalities have emerged in different areas of state action (economy, health, security) when the state gave up on addressing or resolving a given problem (poverty, diseases, criminality) and opted instead for ‘managing’ it by keeping it at an ‘acceptable’ level (Aradau & Van Munster, 2007, pp. 98–99). In the domain of criminal law, this attitude is linked to ‘actuarial justice’, an approach that accepts criminality as a social fact and aims at managing the risk deriving from criminality and avoiding the occurrence of single, specific criminal acts (Yeung, 2018, p. 512). Also this general characterisation of algorithmic regulation as connected to the ‘management’ of risk and crime can be specified and enriched by a specific analysis of the approach to managing risk and individual behaviour within the EU PNR framework.

Indeed, the pre-emptive, risk-based security model of the PNR Directive not only clearly departs from reactive approaches, which address past events only, but also differs significantly from typical preventative approaches, which aim to address the causes of a given phenomenon (criminality, terrorism etc.) and to tackle it in a structural manner. A pre-emptive approach is much more interested in correlations than in causes, and is much more focused on individuals than on structural conditions.

As we have seen, the behaviour potentially leading to a high-risk score typically does not consist in prohibited actions. Additionally, within the EU PNR framework, risk criteria are not publicly known, since, so the argument goes, publicity would make them useless by enabling potential criminals and terrorists to modify their behaviour in order not to match the criteria that lead to a high-risk classification (Ulbricht, 2018, p. 30).

This functioning has critical consequences on the way we understand the relationship between facts and norms and on the role of norms as providing orientation for individual behaviour. Indeed, the risk-based security approach of the PNR Directive seems to be indifferent towards prescriptions for conduct as ‘external’ and more or less stable sets of norms and seems to rely instead on a sort of flexible normativity immanent to the empirical data analysed. Its aspiration is neither to provide rules of behaviour, nor to re-educate or discipline subjects.55 On the contrary, individuals should behave as ‘naturally’ as possible in order for the system to provide accurate predictions. Far from dissuading criminals and terrorists from committing the planned actions (either because of the possible punishment if they transgress laws or because of the possible failure of their plans due to strict checks), these pre-emptive security measures presuppose that potential criminals and terrorists can be more effectively identified (and stopped) if everyone behaves without knowing what actions are considered to indicate a threat.

Yet, behaving ‘naturally’ or ‘normally’ does not ensure that individuals without criminal or terrorist intent will receive a low-risk score, if their behaviour deviates in ways considered to be relevant for risk-profiling from average or ‘logical’ behaviour. It is evident that the ‘norm’ here has nothing to do with external prescriptions of allowed or prohibited behaviour; rather, it simply coincides with the ‘normality’ immanent to the empirical data.

This is particularly evident whenever risk criteria are automatically generated. It is not clear whether such automatic generation of risk indicators (as an alternative to their ‘manual’ definition by LEAs) is already used with the PNR framework, but it is surely part of the risk-based security logic and part of the techniques currently developed as part of the medium- and long-term vision.66 Principally, any kind of behaviour that can be automatically captured and codified into patterns (such as, say, the patterns followed and the time needed to select the purchased itinerary while booking a ticket, travellers’ movement trajectories at border crossing areas, the time spent and the number of visits at the airport’s toilets, and so on) can be analysed against automatically generated risk indicators. The systems can be calibrated to provide alerts on the top 0.1% of travellers whose behavioural patterns deviate from the ‘standard’ trajectories most significantly.

Even if the risk criteria are ‘manually’ set by humans and not automatically generated by machine intelligence, they can be meaningfully used only if they allow for the selection of a small group of travellers. Whether, for instance, paying by cash can be considered a risk indicator or not depends very much on the share of travellers that use other means of payment. If cash payment is the norm on a given route, then the criterion ‘purchase by cash’ becomes useless as a risk indicator. So, the value of a given criterion as a risk indicator, rather than being a socially or legally fixed characteristic of a given action (such as being forbidden or being potentially damaging to others), depends to a high degree on the empirical behaviour of other people in the same situation.

The risk-based logic of security emanating from the EU PNR framework thus consists in shifting the relationship between facts and norms towards a kind of normativity that is directly dependent on the empirical data processed. The norm is ‘mobile’ (Amoore, 2011, p. 31 and 2013, p. 66; Leese, 2014, p. 505) in the sense that the ‘normality’ according to which ‘deviations’ are defined is not given as a fixed standard; rather, it fluctuates. The ‘mobile norm’, or ‘data-informed-code’, as Mireille Hildebrandt has defined it (2018, p. 3), however, does not coincide with the overarching legal norms prohibiting criminality and terrorism and attaching sanctions to them. Rather, the paradoxical normativity of the PNR Directive opens up a space which differs from the legally normed space, which distinguishes acceptable and prohibited behaviour and attaches sanctions to the latter. It locates itself in an intermediate space, the space of the setting of risk criteria, according to which behaviours are sorted and possibly ‘flagged’ as deserving closer inspection. The ‘flag’ shares this intermediate character, since it locates itself between a ‘simple’ recommendation and a sanction (Bellanova and de Goede, 2022, p. 105).

In this space, at the level of risk criteria, the normativity generated by pre-emptive security thus gives up even the aspiration to set norms that are independent of the empirical reality and refer to a more or less external and stable regulatory system. Its main intent is to detect deviations within the Sein (what is), independently of their relationship with a Sollen (what ought to be).

4.1Algorithmic security and the rule of law

Algorithmic regulation, especially in the field of security, has far-reaching implications for the moral, social and legal principles which are highly valued in democratic societies committed to the rule-of-law. Legal scholars have focussed on data-driven decision-making’s impact on the rule of law (Bayamlıoğlu & Leenes, 2018; Hildebrandt, 2018). In connection to the emerging European PNR framework, moreover, Matthias Leese (2014) has highlighted how profiling techniques undermine anti-discrimination safeguards, as well as the principles of proportionality, purpose limitation and accountability. Vagelis Papakonstantinou and Paul De Hert (2015) stressed the vagueness of profiling criteria and lamented the lack of provisions which might ensure effective redress mechanisms in a (at the time they wrote) draft version of the PNR Directive.

In the following sections, I aim to integrate these strains of the debate by highlighting two further kinds of implications for the rule of law deriving from the European PNR-based approach to security. The first of these is concerned with the self-expanding logic of pre-emptive security and its negative impacts on the effectiveness of check and balance mechanisms. The second refers to the impact of individual risk-assessment on legal certainty and individual autonomy.

4.2Existing checks and balances in face of the self-expanding dynamic of pre-emptive security

The pre-emptive PNR logic, has been argued, equates to a securitisation move, according to which every kind of behaviour is susceptible to being captured and analysed for security purposes (Ulbricht, 2018, p. 156; Amoore & Raley, 2016). A look at the European PNR framework shows how this securitisation move, once set in motion, has an immanent tendency to reinforce and perpetuate itself.

First of all, because the aim of pre-emptive security is to identify ‘unknown suspects’, the focus of behavioural profiling cannot be restricted to selected categories of people; it must include the largest sectors of the population it feasibly can. Indeed, the PNR scheme is based on the collection and analysis of data of all air travellers. Each and every person booking a flight is categorised into a risk class based on data collected during the booking and check-in processes, even if most travellers are classified as low-risk in the end. Large amounts of data from people having no criminal or terrorist intent are also needed in order to derive a standard of ‘normal’ or ‘average’ behavioural patterns against which anomalies and outliers can be defined (or used as training material for the automatic generation of risk-criteria).

Moreover, currently there is a growing pressure towards the extension of the categories of PNR data beyond the ones commonly collected during booking and check-in, as well as towards the expansion of the risk-based security logic to further areas. This tendency contradicts the clear commitment, expressed during the preparatory phase and in the PNR Directive, not to impose the collection of any additional data than the ones already collected for commercial purposes (EC, 2011, p. 11; PNR Directive, Nr. 8).

The first PNR review report provides interesting insights that illuminate a tendency towards the expansion of the quality and quantity of data collected and processed. It suggests, for instance, adding passengers’ dates of birth to the compulsory information to be collected and transmitted to the PIUs and to extend the duty to transmit PNR data also to tour operators and travel agencies (EC, 2020a, pp. 10–11; EC, 2020b, p. 43). A possible further development mentioned in the EC Report regards the extension of the collection and use of PNR data beyond air traffic to include rail, ferry and bus transport (EC, 2020a, p. 10).

Pending the aforementioned CJEU decisions, the EC opted not to introduce any changes to the PNR framework at this junction (EC, 2020a, p. 12). Nevertheless, the potential developments mentioned in the review report are indicative of a mental shift from using the ‘data exhaust’ of information already available to the active and purposeful generation of additional data. This resembles the way in which collateral data exploited in the commercial sector started to be ‘hunted aggressively and procured largely through surveillance’ once they proved to be lucrative (Zuboff, 2019, p. 94).

The suggestions included in the first review also indicate what will be on the agenda if the CJEU decision leaves room for further developments, or if unexpected events will once more create a political climate favourable to the acceptance of extended surveillance measures. Especially the broadening to rail, ferry and bus traffic would entail an expansion of surveillance opportunities and apparatuses. Indeed, it would adapt these modes of transport to the booking and check-in procedures we are accustomed to for flying, including compulsory nominal reservation and pre-boarding checks. If the CJEU Decision will follow the AG Opinion, indeed, no significant and generalised limitations to the collection and processing of passengers’ data are to be expected. Although the Opinion recommend to limit the generalised retention of data in a non-anonymised form, it indeed considers ‘the transfer and the generalised and undifferentiated automated processing of PNR data […] compatible with the fundamental rights to respect for private life and to the protection of personal data’ (CJEU, 2022, p. 1).

A further crucial aspect of the self-expansive logic of risk-based security derives from the use of non-criminal data for security purposes. The non-criminal data on which the PNR framework relies for the risk-assessment consist largely in non-verified information. These data are not taken from official databases, verified by security operators or otherwise checked for accuracy, but are based largely on information provided by the travellers themselves, for instance when they book a ticket online. This means a potentially high level of inaccurate information entering the PNR database, a correspondingly high rate of mismatches with official databases (due for instance to errors in spelling, typos, etc.) and additional difficulties in drawing reliable results for the risk assessment. It is difficult to imagine how this inaccuracy could be counteracted if not by activating a verification spiral which intrudes more and more into further spheres of human activity. Examples of such verifications include required uploading of an ID card and biometric real-time online identification during the booking process in order to make sure that the ID card matches the traveller and the verification of phone contact details via real-time messages and calls. Such practices are already realities in commercial applications, and an uptake by the security domain seems to be a logical development along the path of risk-based security inspired by commercial practices.

The pre-emptive shift of security measures and their self-expansive character poses serious challenges for traditional check and balance mechanisms typical of the rule of law. On the one hand, ex-ante control mechanisms such as parliamentary authorisation are at risk of being trivialised. The pre-emptive logic postpones the justification of security measures to a future that is not verifiable. The problem which these measures should solve are not present (an increase in criminality or terrorist activity) but are projected into the future (a possible terrorist attack, a possible smuggling of a big quantity of drugs). Thus, the justification of these measures and of their continue expansion is also projected into a possible future and can neither be verified nor invalidated. Reference to past empirical evidence can hardly be used for discussing the necessity and proportionality of the suggested measures. Moreover, being the justification of pre-emptive security measures built on the avoidance of single, exemplary criminal episodes and terrorist attacks, ex-ante checks can be easily circumvented under the pressure of particularly striking crimes or terrorist attacks, as was the case when the PNR Directive was passed by the EU parliament, after years of reluctance, following the Paris terrorist attacks.

On the other hand, traditional ex-post checks, such as judicial reviews, though still as important as parliamentary authorisations, are episodic and reactive by nature. Their activation is bound to high procedural thresholds and preconditions and therefore can take place only in select cases (TEU, Art. 19, TFEU 251–281). They risk becoming effective only years after a questionable measure has been implemented. This holds not only in the case of the pending CJEU decisions on the compatibility of the PNR Directive with fundamental rights, which, in the most optimistic case, will be taken six years after the directive’s entry into force, but was also the case of the Data Retention Directive, which the CJEU declared to be incompatible with EU fundamental rights in 2014, eight years after its adoption.

4.3Legal certainty and individual autonomy vs the paradoxical normativity of risk-based security

The existence of a relatively stable and known set of rules providing orientation for one’s behaviour is a core characteristic of (modern European) normative systems. Since the political turn of the modern era that has equated legitimate power with power tamed by the rule of law, the principle of legal certainty has played a crucial normative role in constraining arbitrary power. Legal certainty, according to its general meaning, requires that ‘the addressees of laws must know the law in order to be able to plan their actions in accordance with it’ (Raitio, 2020).

Although the principle of legal certainty, on a strict, technical legal interpretation, applies specifically to criminal law and judicial decisions, it applies in its broader sense also to decision-making in the hybrid domain covered by the PNR framework, which, as discussed above, opens up a hybrid space of normativity and behavioural influence between border controls, mobility surveillance and more classical law enforcement measures.

Within this space, individuals are subject to decisions by public authorities (i.e. perform closer scrutiny, prevent from boarding the airplane or crossing the border) based on opaque criteria. The addressees of these measures and norms are not provided with elements to guide their own actions.

This lack of transparency poses a problem and a challenge for the rule of law itself, as has been stressed by Yeung (2018, p. 517) and Hildebrandt (2016). Additionally, it implies a fundamental shift not only in the way the relationship between the ruled and the rulers is constructed and reflected through the regulatory framework, but also in the way individuals as addressees of the norms are conceived.

Bayamlıoğlu and Leenes (2018) elaborated on a similar aspect with specific focus on algorithmic regulation in the judicial context. They see intelligibility, reliability and predictability of the normative order as the core constituents of the rule of law. In traditional ‘code-driven’ regulatory systems, these attributes of the normative order are realised through the ‘normative force’ of the law:

‘Data-driven’ (or ‘algorithmic’) regulatory measures instead reject this causal enterprise of (explicitly and transparently) connecting legal facts with legal consequences and thus lack the normative force which is characteristic of regulatory measures under the rule of law. For Bayamlıoğlu and Leenes (2018, p. 309) this lack is problematic, because it constitutes a blow to human autonomy and deprives individuals of their right to effectively contest decisions by public authorities which affect their lives. Bayamlıoğlu and Leenes are principally concerned with the ineffectiveness of redress measures. Although Art. 15 of the PNR Directive establishes that national supervisory authorities must deal with complaints by individuals, the opacity of the risk criteria (and possibly of the subsequent measures) negatively affects the effective exercise of complaint rights. Bayamlıoğlu and Leenes’ observations, thus, seem to apply to the EU PNR regime as well.

However, in my view the demise of the causal enterprise of regulatory measures operated by the PNR Directive additionally poses a problem for individual autonomy in a more fundamental way. The European PNR framework, by using non-criminalistic, mundane behavioural data, negates the possibility for individuals to direct their own behaviour based on predictions of (legal or administrative) consequences. It thereby challenges the conception of individuals as moral agents who are able to modify their own behaviour by orienting their actions on the basis of general and abstract norms. This problem in relation to human autonomy emerges prior to the stage in which redress mechanisms can be activated, when individuals can challenge decisions and measures taken by authorities.