Shipwreck after Shipwreck

On 26 February 2024, the European Ombudsman issued a decision OI/3/2023/MHZ on the fundamental rights obligations of the European Border and Coast Guard Agency (Frontex) with regard to search and rescue (SAR) in the context of its maritime surveillance activities. In particular, the decision concerns the Adriana shipwreck of 14 June 2023, also known as the Pylos shipwreck. The investigation was initiated shortly after the human tragedy off the Greek coast of Pylos, where over 600 migrants lost their lives, amid concerns pertaining to Frontex’s role in maritime incidents and, more broadly, the continuing loss of life in the Mediterranean Sea.

While affirming Frontex’s compliance with the applicable rules and protocols, the inquiry exposed significant shortcomings in how the Agency handles maritime incidents, including the issuance of emergency signals, both in the context of joint maritime operations and its separate multipurpose aerial surveillance activities. Given the persistent scale of recurrent shipwrecks, I argue that integrating AI systems into Frontex’s activities, which are informed by a normative framework for the classification of maritime emergencies, has the capacity to significantly improve the decision-making process in responding to boats in potential distress and the overall SAR system. Such development in this area will be aimed at enhancing the protection of life at sea, provided that the flaws inherent in human assessments of cognitive biases and response time delays are adequately addressed.

The Legal Framework on Search and Rescue and Emergency Situations at Sea

The 1979 International Convention on Maritime Search and Rescue (SAR Convention) sets the framework for the international SAR system, including the responsibilities of coastal States, and provides the definition of a “distress phase” as a “situation wherein there is a reasonable certainty that a person, vessel, or other craft is threatened by grave and imminent danger and requires immediate assistance” [Annex, para 1.3.13, see also, the analysis here]. The law also delineates the criteria for classifying the phase of emergency of boats detected at sea and outlines the circumstances under which MAYDAY alerts and other emergency signals should be issued. Despite this, the scrutiny of these additional emergency signals, which could prove tantamount to safeguarding life, often escapes investigative attention and practice, leading to gaps in Frontex’s response strategies.

Emergency Alerts: Distress Signal, Urgency Signal, and Safety Signal

Particularly, Frontex has the capability to issue a MAYDAY alert to all vessels nearby – a distress signal signifying imminent threat to life requiring the most urgent response – in the course of its (joint or individual) search activities when it observes an emergency. Conversely, two alternative signals may be issued: a) a PAN-PAN, which indicates an urgency signal pointing that a vessel is in jeopardy but does not imply an immediate threat and has priority over all but distress traffic, and b) a Sécurité, which indicates a safety signal for messages concerning the safety of navigation or giving meteorological warnings to the Rescue Coordination Centres (RCC).

Interestingly, neither of these alternative emergency signals has undergone an examination in the present inquiry, as Frontex was found not in breach of the relevant rules and provisions. This implies that emergencies are perceived as either involving a grave and imminent danger to life or vessel, or not at all – no room for in-betweens. In this regard, the European Ombudsman recommended that Frontex ensure a higher degree of independence and that “the provisions on aerial surveillance in operational plans and in-service requests under EUROSUR, while compliant with law, enable it to act, to the greatest extent possible, on the recommendations to maintain surveillance at ongoing maritime emergencies” [at 55]. The fact that Frontex, in the present case, offered on various occasions to assist the Greek RCC by providing surveillance but received no response, was deemed sufficient for the purposes of showing “effort” to follow-up. Practically, its units remained absent from the scene leading up to the shipwreck, even though they should have continued to monitor the situation unless “otherwise dictated by force majeure” (Frontex report here, albeit it was taken down on 28.02.204).

When should life or vessel be considered in the distress phase? Adopting the United Nations High Commissioner for Human Rights’s broad understanding of distress, I posit that the notion of distress should encompass all “unseaworthy vessels even if they are not in imminent danger of sinking” (see for similar views here and General Comment No.36 of the UN Human Rights Committee on the right to life). The historical context of unreliable vessels used to transport migrants in the Mediterranean region lends support to the wider inclusion of overcrowded vessels under the distress phase as it indicates unseaworthiness and, hence, a serious threat to human life. In this vein, the EU Sea Borders Regulation requires Frontex operations to take into account “the seaworthiness of the vessel and the likelihood that the vessel will not reach its final destination” when determining if there is a distress situation. This is instructive in developing the normative content of emergency phases, as opting for a narrow construction of the term “distress” might lead to an irreversible violation of the right to life. Manifestly, Frontex’s interpretation of the Adriana was strictly insular, as it failed to classify it as “distress” and relay a MAYDAY alert despite its aircraft showing that it was “heavily overcrowded and was navigating at slow speed”.

Furthermore, Frontex’s assertion that it bears no responsibility and lacks the competence to designate the phase of emergency (uncertainty, alert, and distress) is also troubling as it contradicts current practice. On various occasions in the past, the Agency had relayed MAYDAY alerts on the basis of a boat being in a “severe state of non-sea worthiness and about to sink” or “where there are already people in the water” or where it believes that the relevant RCC is not aware of the incident or not already coordinating a response [at 33, 58]. At this point, we might recall the early judgment of The Eleanor dating back to the 19th century, where the English High Court of Admiralty attached to the term “distress” an implicit need for urgency but that “there need not be immediate physical necessity” (para. 135). Consequently, it remains unclear how “active” the role of Frontex’s Team Leader is in relaying MAYDAY alerts or the alternative emergency alerts. As such, it appears that the designation of emergency state is only done incidentally. Still, the European Ombudsman maintained that the exclusive SAR responsibility to launch, organize and coordinate a SAR response remains with the relevant RCC [at 48, 62]. In order to strengthen communication between the RCCs and Frontex, it is primarily fundamental to understand the context of each sighting. This could arguably best be achieved through the incorporation of AI to assist human-agents in the decision-making process when identifying a potential emergency.

Objective vs Subjective Judgment: Machines and Human Decisions

The failures to designate emergency situations, as emphasized by the European Ombudsman, may in fact be attributed to the absence of internal guidelines that will be essential in situations that are “extremely stressful” to detect such emergencies, as “the staff concerned witnessing potentially life-threatening situations” know “that they cannot have a definitive say on the ultimate response to the emergency” [emphasis added, at 65]. Emotional decisions in stressful circumstances might as well lead to irrational outcomes or result in less objective designation of an emergency, influenced more by human bias. However, emotion can also be perceived as a positive factor in showing empathy in such situations. Drawing parallels with criminal court trials, where humans are often subconsciously influenced by emotion and preferences and the judge asks the jury to set aside their biases (emotion) in coming to a decision, the argument lends support towards the view that machines are more adept at making or assisting in decisions than humans. Machines have therefore the “advantage” of not being vulnerable inter alia to stress, frustration, and emotion and may thus be perceived as more impartial in making a decision to issue MAYDAY alerts based on normative criteria and risk factors. Notwithstanding, the issuance of an emergency signal does not guarantee that the responsible RCC will launch a SAR operation [at 61] but provides an objective way of overcoming human subjectivity and tantamount support that appropriate measures should be taken for the overall coordination of SAR services according to existing plans of operation (SAR Convention, Annex, para 4.5.3 and para 4.1.3).

No Tolerance for Avoidable Shipwrecks

The need for Frontex to reflect on how to approach “the issue of emergency signals for similar incidents in the future” was also highlighted as a shortcoming of its practice by Frontex’s Fundamental Rights Office. Undoubtedly, AI systems have the potential to significantly enhance the decision-making process in this context, as they can process and be trained to vast amounts of real-time relevant data, including weather conditions, vessel type, thermal images and historical incident data. Advanced analytics in this area can assist in identifying patterns and anomalies, helping to distinguish between the different phases of emergencies based on existing judicial guidance and the factors included in Article 9(2)(f) of EU Regulation 656/2014. In a futuristic vein, the development of this surveillance system will make an assessment and predict when human life or vessel is, or will be, in imminent danger – before a shipwreck takes place – and recommend proactive action, including the timely and effective broadcast of the MAYDAY alert, which would then allow the RCCs to respond more efficiently according to the emergency phase of each factual scenario. A counter argument that requires further consideration is the possibility of false positives, which could create confusion, especially in the case of multiple distress incidents and the need for concurrent SAR operations.

An analogy can be drawn here from the events in the Crotone shipwreck, which are also discussed in the Ombudsman’s decision. The Frontex aircraft that detected the boat during its surveillance activities did not classify it as a potential emergency situation, while its own aircraft had to return to base due to deteriorating weather conditions. This indicates that the detection of a wooden, overloaded boat in prevailing or forecasted extreme weather and sea conditions, along with significant thermal imagery from below deck, does not warrant an emergency classification. Paradoxically, the boat was treated to be seaworthy, contradicting established criteria and Frontex’s SAR duties under Regulation (EU) 2019/1896, which the Frontex’s Team Leader should take into account when designating emergency situations. Moreover, below the threshold of distress, legal duties of Frontex and RCCs also arise from the uncertainty or alert phase (see here).

The history of casualties in the Mediterranean, shipwreck after shipwreck, underscore the urgent need for more rapid, impartial and objective responses to situations when life is in danger of being lost. For Frontex to adhere to its professed capabilities, that is, contributing inter alia to a comprehensive situational picture in the Mediterranean and “at the same time saving lives thanks to [its] real-time monitoring and quick reaction capabilities”, it needs to take proactive measures to address the systemic issues in its operational “search” practice. Considering the expanding usage of AI mechanisms in migration, asylum and border control management, coupled with the EU’s substantial spending on research and development for security technology (annually over a billion), I argue that it is critical to harness existing technological tools in line with international obligations to better respond to emergency situations and minimize the loss of life at sea. In parallel, it needs to be underscored that the EU AI Act considers border technologies of this type as high-risk, which implies that necessary safeguards such as transparency obligations, human rights impact assessments and scrutiny, are crucial. These measures are tantamount if any such development is to enhance protection of the rights of refugees and migrants and not exacerbate the problematic status quo, namely, violence and practices of remote control that facilitate interceptions and returns (see also, here).

Conclusion

The recommendations of the European Ombudsman, stressing the need for Frontex to explicitly lay down guidance for its staff as to when MAYDAY relays should be issued under both the context of its joint operations and within the framework of EUROSUR aerial surveillance [at 64], are particularly helpful and much needed. Shifting focus from a dehumanizing border regime to investing in beneficial technology has the ability to reinforce the Union’s rhetoric around saving lives at sea in a practical and real manner. Ethical considerations are tantamount in exploring this option, as human oversight in decision-making is crucial in maintaining accountability and ethical standards. More broadly, for AI to fulfil this promise, any algorithm developed should be primarily tested with pilots and be subject to continuous assessments, especially when applied in the context of life and death decisions. Admittedly, the intricate role of human judgment remains one of the most complex philosophical questions.