BorderForce promises “a new era of border surveillance”. According to a reply the REA sent to our inquiries, the project aims to realize it through the development of enhanced “real-time surveillance capabilities using AI”, which will allow border authorities to quickly respond “to dynamic and evolving threats and challenges by versatile means including the use of anti-drone capabilities, satellite and UAV assets, autonomous sensors and relevant OSINT”.
The assumption behind this alleged paradigm shift is that while fixed surveillance assemblages already cover 13% of the EU’s external land borders, an adaptable, flexible, “dynamic system” that is capable of responding to “evolving threats” is urgently needed.
This is where the BorderForce system comes in. While stating a specific (but rather vague) emphasis on “ethical, legal, and social aspects, safeguarding fundamental rights in border surveillance capability development”, it aims to ensure “seamless operations in monitoring the flow of goods, people, and information” at the border, addressing not just migration and smuggling but also “geopolitical tensions”. As such, “BorderForce contributes to regional stability, particularly in crises”, writes its Cordis page.
Interestingly, BorderForce is seen as a model unit that can — and should — be replicated at borders all over the globe: “The BorderForce solution for monitoring people, flow of good and border relevant information will be replicable across European external land borders and shorelines as well as third countries in the context of CSDP (Common Security And Defence Policy, ndr) missions”.
From the redacted Grant Agreement we obtained through an access to information request, we also learn that “the overall aim” of the BorderForce project “is to achieve a commonly accepted framework and system for internal and external border surveillance” — which means that the vision includes the monitoring of borders between Member States, in case of “temporary” suspensions of Schengen.
Even though the project’s solutions are admittedly tailored to the specific, operational needs of the security practitioners, the project boasts “a socio-technical innovation-inspired approach”.
According to the redacted Grant Agreement we obtained, BorderForce aims to develop “a scalable and deployable real-time threat assessment system to automatically provide geo-referenced threat indicators based on autonomous and adaptable smart C2 (Command and Control, ndr) station for border surveillance integrating versatile sensing platforms and OSINT information”.
This translates into the ambition to go beyond the state of the art in:
1) “UAV-based and Satellite-based monitoring”. “The use of UAS in law enforcement has exponentially grown in recent years, including border management”, but “despite widespread use, the performance of standard UAVs remains constrained for example by limited monitoring and communication distances, high energy consumption, weather conditions and limited night-time operations”. Similar issues affect current satellite-based monitoring techniques.
2) OSINT (i.e., extracting information from publicly available sources, e.g. social media): “reported uses of OSINT still remain scarce in border management addressing primarily border checks applications, not border surveillance. OSINT along with various other data could however facilitate proactive response towards detected, tracked and identified high-risk entities and overall provide significant support in generating comprehensive situational awareness”. In other words, BorderForce aims to use analyses of social media and other publicly available online content of high-risk individuals to proactively (or better, predictively) intervene against them even before they can act/pose an actual threat.
3) Data fusion, as “further improvements are required” on the accuracy of sensors from which data are collected and cross-referenced, which might not be optimal if — as in most cases — these are not tailored for specific border surveillance needs. Social media are evoked here as well, as “The fusion of sensor analytics and OSINT analytics represents a prominent research topic in this area”.
4) (Augmented) Situational Awareness, which aims to deploy “extended reality” systems (XR) to “significantly reduce training costs by enabling decision making based on provided situational awareness”. XR could even give drones our own eyes: “Mixed Reality has not been previously utilized in the field of border security to visualize camera views or drones’ coverages. Therefore, the combination of mixed reality with exact outdoor localization is expected to have a tremendous impact in discovering non-covered areas on-site and efficiently guiding responses to localized threats”.
5) Data connectivity, which should be strengthened in remote/rural areas.
6) Architecture and cyber-security of border surveillance systems, which currently “often come with a limited range and coverage” and rely on “fixed cameras or sensors”. Automation is seen as key here, as legacy systems have been “heavily reliant on manual monitoring”, and are therefore “prone to human factors such as fatigue or distraction”.
Technological capabilities are more precisely detailed in an article on the March-April 2025 edition of the Border Security Report journal. These include:
1) “self-sufficient, transportable Command and Control (C2) Stations with configurable and extendable capabilities”, integrating “AI-based detection for autonomous optical edge sensing and (anti-) UAV functionalities”, the “target objects” being “persons, vehicles, animals, vessels, and UAVs”;
2) “versatile surveillance towers with anti-drone features, integrating data from autonomous monitoring sensors and UAV systems”;
3) satellite resources, such as “Low Earth Orbit (LEO) satellites, Cubesats with high revisit times, and Copernicus satellites, along with Copernicus Contributing Missions (CCM) to gather crucial data”;
4) “drone swarm sensing with sophisticated flight plan updates” (which, according to the project website, feature “Real-time tracking and object geo-location”, “Automatic detection and classification of suspicious objects within a 5 km² area” and “4K RGB cameras and thermal cameras for detailed reconnaissance, day and night (minimum 1080p)”);
5) “an OpenSource Intelligence (OSINT) platform”, to “derive threat indicators relevant to border regions”. According to the project website, it will also include “media intelligence” features, such as: AI image detection, image manipulation detection, fake face detection, morphed face detection, deep fake detection, and even “geolocation estimation”, allowing the authorities to “estimate GPS coordinates from image data” (detailing “Location on earth where image motive could be taken: Latitude Longitude”);
6) “an autonomous UAV-aided mesh wireless communication network, managed through RPAS and VTX Mesh, enabling seamless data exchange in challenging environments”
7) data fusion, which is “carried out on features extracted from sensor data and the OSINT sub-system”. As the project website further specifies, “The platform applies predictive analytics to classify potential risks by type and severity. It also tracks unfolding incidents in real time and employs explainable AI to identify and highlight the key contributing features”, informing a risk assessment component “that envisions to leverage FRONTEX’s Common Integrated Risk Analysis Model (CIRAM) to ensure a consistent and standardised evaluation of threats across the system”. This will also generate “georeferenced Risk Indicators by integrating fused sensor data with potential features derived from Open-Source Intelligence (OSINT), maintaining geolocation metadata from the original sources to preserve spatial relevance and traceability.”
8) XR features that promise to be mobile and include the augmented monitoring of a 3D map of the area, together with the interaction “with real-time XR elements like movable observation towers, alarms, sensor data, UxV videos, camera streams, and other tactical information”.
9) Advanced, multi-sensors anti-drone solutions for the “early detection and mitigation of airborne threats”, including features such as radio frequency and “cyber-RF” sensors (for “controlled takeover” of the hostile UxV), jammers, optical and thermal cameras. These will be developed by Securiton Germany.
Lastly, BorderForce is meant “for land border surveillance across small (200m) and larger areas (up to 20km), addressing diverse environmental conditions and weather scenarios”, concludes the Border Security Report article.
The project builds on I-SEAMORE and EURMARS, as requested by its funding call. FOLDOUT is also mentioned (unsurprisingly, given that the project coordinator, AIT, also coordinated FOLDOUT) as an example of — much needed, the Annex of the GA we received alleges — “research to enhance the capabilities of basic surveillance towers by integrating multi-modal sensor technology to standard surveillance cameras and supporting human operators with AI-based event and object detection”.
In fact, FOLDOUT vowed to address issues encountered by smart sensors and advanced analytics in difficult weather and vegetation conditions. This is seen in BorderForce as crucial to go beyond the state of the art in “mobile and deployable border surveillance solutions” — basically AI-based stations that, instead of being fixed and providing “only rudimentary automated event detection functionalities”, can be transported and adapted according to needs.
The Austrian MOBILIZE project —which featured the AIT as a member of its Consortium as well — is also mentioned, as “in the MOBILIZE project, the operational and technical concepts for temporarily deployable and relocatable technical security systems for large-scale railway systems” were “examined and demonstrated”.
Crucially, and miraculously, we could read a non-redacted section dedicated to ‘Linked research and innovation activities’ in the GA we obtained.
It includes a fundamental “List of key projects related to BorderForce and how BorderForce exploits them’. This features several projects we investigated (as it “builds upon, reuses, and integrates outcomes of recent BM projects” such as EURMARS, FOLDOUT, CRiTERIA, PROMENADE, NESTOR, ARESIBO, ROBORDER and TRESSPASS) and even lists the specific components which will be exploited by BorderForce.
These include:
– the “fusion approach of real-time border surveillance data combined with web and social media information” developed in NESTOR
– the “dynamic risk assessment system module” developed in project TRESSPASS
– multisensor systems developed in EURMARS and FOLDOUT
– “movable situational awareness tool for tactical commanders using XR” developed in ARESIBO
– methodologies (for requirements development and concept of operations [CONOPS]) developed in ROBORDER.
Lastly, Consortium members include HARDWARE AND SOFTWARE ENGINEERING EPE (GRE), which also participated in the EURMARS project, and TEKNOLOGIAN TUTKIMUSKESKUS VTT OY (FIN), which was also part of EURMARS, FOLDOUT, ROBORDER, and D4FLY.
In the redacted Grant Agreement we obtained, we could read that the objective is to bring the prototyped BorderForce solution(s) to Technology Readiness Level 7 (TRL7), “to enable smooth and unproblematic execution of tests in an operational environment”. According to Horizon guidelines TRL7, in fact, amounts to a “system prototype demonstration in operational environment”. In other words, and more concretely, “Reaching to this point, one can conclude that the new technology is reliable from the technological point of view”.
Part B of the Annex included in the document we obtained confirms: the project is divided into two phases, the first in an “industrial environment” to obtain a TRL5 solution (meaning that the technology is validated in a relevant environment, or more concretely that “Reaching to this point, one can conclude that the new technology is feasible from a technological point of view”); the second one in an “operational environment targeting TRL 7 for the final BorderForce system”.
Overall, this means that the developed solutions will not immediately be ready for the market — but could soon be.
Operational assessment will be performed during two (as yet unspecified) field trials. We only know that “Since BorderForce is envisioned to develop a wide range of AI-based systems, the BorderForce consortium is dedicated to adhering to stringent measures and procedures, ensuring strict compliance with the directives governing the use of AI systems”.
Several crucial project details are redacted in the Grant Agreement we obtained from the EU Commission’s Research Executive Agency, including “project outcomes”, “expected impacts”, the exploitation plan (of which we could however read that it includes a “go-to-market strategy”, with a “market launch phase” to be finalised at the end of the project), the “security” section and an interesting “justification of subcontracting” section of which we were not able to make sense.
Detailed descriptions — and in some cases, even the shorter “objectives” — of each individual Working Package were also fully and consistently blacked out in the initial release, and only very partially disclosed after we challenged its rationale with a formal request (“confirmatory application”) to the Director of the REA, Marc Tachelet.
In fact, “The project deals with security-related information, research and operational methodologies, some of which involve confidential strategies used by public authorities. Their disclosure”, argued Tachelet, “would undermine public security. Certain sections of the document describe technological vulnerabilities or operational procedures that, if disclosed, could be misused by malicious actors.”
From a mostly redacted slide, we could however read that having “No direct connection between intel from the digital world (social media) and illegal activity on the ground” was one of the “specific needs” of security practitioners that “triggered” the BorderForce project.
It is a matter of concern that, as visible in the redacted GA, the Border Guards of Albania, Slovakia, Moldova, Romania, Lithuania and Bulgaria led the development of the BorderForce proposal, according to their own needs — as defined by themselves — and use cases. “Close cooperation between BGs and developers is a key success factor”, reads the document. This cooperation is envisioned to last from the very beginning of the project to its completion.
Not exactly a “co-design” procedure that involves all stakeholders, giving each the same voice.
And yet, BorderForce boasts “the co-creation of knowledge among people from several fields with different backgrounds crossing traditional boundaries between different disciplines”, and therefore claims to be “by default interdisciplinary”.
It also boasts integration of social sciences and humanities (SSH), “to responsibly develop tools with effective mitigation measures against adverse consequences for society”. In particular, “Strong SSH engagement is intended to mitigate the risk of a Collingridge dilemma: increase understanding of potential impacts before the design of the technology has been solidified and negative impacts have become harder to address”.
When it comes to ethics, after the usual commitment “to adhering to the highest ethical standards in research”, the Annex writes that “the project may raise ethical considerations in three categories: human beings, personal data, artificial intelligence”.
An “unwavering commitment” is stated to “legal compliance, ethical desirability, and social acceptability”.
“Extensive research” will also be conducted during BorderForce “on the ethical and legal issues entwined with its technologies and methodologies, with a particular focus on ethics issues, risks identification, and the development of mitigation strategies”. The project website further specifies that “an Ethics Assessment will be performed internally throughout the project by the Partner Eticas”.
More specifically to its AI models and decision-making processes, the project ensures that they will adhere to “human-centered values, fairness, transparency, and explainability” as enshrined in the OECD’s “Recommendations on Artificial Intelligence” and the Council of Europe’s “Study on the human rights dimensions of algorithms“. The AI Act is not mentioned in this context, even though we managed to learn that deliverable D2.4 will consist of an “AI Act Compliance Report”.
Lastly, when it comes to going beyond the state of the art in protecting human rights, the language — which was until then brimming with details of surveillance and security technologies — immediately becomes vague in the GA. For example: “While technologies can enhance security and immigration control, there is a pressing need to balance the safeguarding of public safety with respecting individual rights”.
We did learn however, by reading the heavily redacted version of deliverable D2.1 ‘User requirements and system architecture’ we obtained from the REA, that BorderForce is considered a “high-risk AI system for border surveillance”. As such, its compliance with the AI Act implies several specific requirements:
a) “A continuous risk management process (AI Act Art. 9) must be in place to identify, analyse, and mitigate potential harms”;
b) “BorderForce must be trained on diverse data reflecting various ethnicities, genders, and environments to avoid biased performance”;
c) “BorderForce must provide its operators (border agents) with clear instructions and information about how the AI works, its limitations, and the meaning of its outputs”;
d) “The AI Act requires appropriate human oversight measures (Art. 14). BorderForce should be deployed as a decision-support system, not a fully autonomous decision maker.”
This is also consistent with what the GDPR requires, according to the same deliverable: “Automated decisions and human oversight are also critical, as GDPR Article 22 discourages decisions with legal or significant effects based solely on automated processing. In the border context, this means BorderForce’s alerts or risk scores cannot be the sole basis for denying entry or flagging someone without human intervention. A human officer should review and make the final decision, ensuring the person can contest or obtain an explanation for any automated assessment”.
Lastly, the project’s ‘Evaluation Summary Report‘ we obtained from the REA generally provides a very positive assessment of the BorderForce proposal. It however also raises some issues:
– “some KPIs (…) are not sufficiently specific in relation to describing the exact systems, standards, and sensors required to validate achievement of relevant objectives”;
– “while planned implementation of OSINT is original, the proposed path to achieve progress is not sufficiently well described”;
– “the methods and the training of the AI are not described in sufficient detail”; and, crucially,
– “while the proposal mentions a potential contribution to improved border crossing experience for travelers, this is less convincing, as it is not a focus of the proposal.”
