Since Russia began its invasion of Ukraine, the United States has invested over $27 billion in security aid to support Ukrainian forces. This aid has included man-portable air defense systems (MANPADS), anti-tank guided missiles (ATGMs), HIMARS launchers, M777 howitzers, multiple launch rocket system (MLRS) rockets, drones, armored vehicles, and small arms. This support is among the U.S.’s largest annual sums of foreign security aid, surpassing the highest provisions in recent memory, including Iraq ($6.9 billion in 2006) and Afghanistan ($11.9 billion in 2011). Across the political aisle, both Republicans and Democrats have called for greater accountability and transparency for this aid. But this kind of oversight requires insights on the use and flow of foreign weapons to the conflict. As the war persists, the open question remains: How can the U.S. develop a sustainable and secure long-term strategy for sending weapons to Ukraine?
The speed and scale of the investment presented several critical hurdles for properly monitoring the use (and misuse) of weapons in Ukraine. Successful and comprehensive end-use monitoring is vital to the region’s long-term stability but is currently impeded by the conflict’s high intensity, preventing standard, in-person inspections of weapons storage facilities. A leaked cable from the Department of State stressed the security difficulties hampering the movement of U.S. officials in-country. Megan Reed, a spokesperson for the Department of Defense, publicly emphasized this challenge, stating, “End-use monitoring was designed for peacetime.” While monitoring may be difficult, it’s not impossible—fighters and civilian journalists have uploaded countless hours of footage and images of the conflict to social media platforms, documenting the flow of arms into and within Ukraine. Bolstered by machine-learning tools, the U.S. should extract insights from social media data and develop a crowdsourced reporting program to track its weapons in Ukraine.
The risk of illicit arms trafficking will increase as the war progresses. Unfortunately, the U.S. is facing several difficulties for conventional end-use monitoring, compounded by the ongoing intensity of the conflict and the rate, scale, and breadth of weapons transfers. In instances where security aid is provided to foreign governments, the Arms Export Control Act requires that the U.S. implement end-use monitoring programs, introducing oversight measures to ensure partners are complying with international law and safeguarding high-risk weapons. Traditionally, a minimum standard for end-use monitoring involves “conducting timely physical security checks of the storage facilities and inventories” of partner nations. One estimate in November 2022 found that U.S. monitors had directly inspected only 10 percent of high-risk weapons sent to Ukraine. The U.S. has acknowledged that it must undertake additional measures to ensure that military aid reaches the hands of those most in need rather than falling prey to Russian forces or war profiteers. Recently, U.S. provision of security aid has endured numerous incidents of misuse and misplacement. For example, in 2019, several Javelin anti-tank missiles intended for French forces ended up in the hands of Libyan rebels opposing the U.N.-backed government.
Since the onset of the conflict, the U.S. has been able to confirm initial weapons receipt in Ukrainian hands, but the path to the frontlines is less clear after that. Complicating this picture is the complex array of individual units and volunteer militias that not only receive centralized support through the Ukrainian military but also individually solicit funds and arms through online and interpersonal networks.
In a plan released at the end of October 2022, the State Department introduced a three-pronged monitoring strategy. This included (a) bolstering accountability by improving the capacities of existing in-country U.S. support staff to conduct on-the-ground monitoring, (b) strengthening border protection by investing in stricter safeguards to prevent weapons trafficking to neighboring countries, and (c) building capacity by introducing training for Ukrainian security forces to improve recognition of specific, high-risk weapons and to conduct end-use monitoring themselves. In addition, the department stated that they would resume a degree of on-site inspection, depending on security conditions.
However, this plan does not create an understanding of the current state of U.S. weapons use, movement, and theft in Ukraine. These recommendations are also buttressed by the real concern of endangering U.S. officials when conducting in-person monitoring. If the U.S. government hopes to retain domestic support for its efforts in Ukraine and ensure that weapons are not diverted to other conflict contexts, it must develop robust and innovative monitoring mechanisms.
The Value of OSINT for Tracking Weapons
Open-source intelligence (OSINT) methods, which involve using publicly available information to develop actionable insights, offer a rich and well-tested toolkit for the U.S. government to create a baseline data set to improve understanding of the location and use of weapons in Ukraine. Since the conflict in Syria (often referred to as the first social media war), researchers and activists alike have been able to glean unprecedented, real-time insights from the ground in conflict settings, piercing through the fog of war. Much of the Russian invasion of Ukraine has been broadcast online, with civilian journalists uploading photos and videos highlighting the conflict’s human costs in real time. The last year of conflict has generated over 10 years’ worth of recorded video data alone. While many organizations and hobbyists used OSINT to generate new and important insights on the conflict, none explicitly focus on the analysis of U.S. weapons flows into and throughout Ukraine. Without a concerted and technical investment, the U.S. stands to miss out on the potential of this data source to address the current limitations of standard end-use monitoring.
The U.S. can further stimulate a new stream of OSINT data by incentivizing individual fighters and civilians to directly share their images and videos of foreign weapons with relevant monitoring agencies. Images could be uploaded directly or emailed to a secure server, modeled after prior programs aimed at safely soliciting civilian intelligence in active conflict settings, including civilian tiplines. For example, during the Syrian conflict, civilians sent emails and WhatsApp messages to the State Department documenting cease-fire violations. Videos and images often accompanied these claims. Relatedly, Syrian rebel groups have been documented uploading YouTube videos thanking private investors for their support and showcasing their receipt of donated weapons systems. These kinds of videos were often a stipulated requirement for future support.
The State Department could then tap into this rich data source to generate insights into the flow of arms in Ukraine. There’s precedent here. OSINT already has proven applications for end-use monitoring. For example, the date of Russia’s invasion was predicted by research groups and private-sector intelligence units scanning through social media, publicly available information, and satellite imagery for early indicators. The open-source investigative organization Bellingcat has also tracked the use of cluster munitions by Russian forces, identifying their use in specific cities through images and videos posted online.
Collecting Relevant Data on U.S. Weapons
Incentivizing fighters and members of civil society to create a new data set is critical, but the U.S. government can tap into existing ones. Journalists and civil society are acting as aggregators of online data on Russian and Ukrainian activities. For example, the Twitter account @UAWeapons purports to track the geographic location of foreign weapons sent to the Ukrainian military. Its recent posts include images of weaponry with serial numbers and locations, including artillery ammunition from Pakistan, a surface-to-air missile system (9K33 Oska) from Poland, and an armored combat support vehicle (LAV 6) from Canada. To capitalize on this work, the U.S. should dedicate resources to realizing the potential insights of this data source at scale and with the explicit goal of supporting end-use monitoring.
The Facebook and Twitter accounts of Ukrainian military units and militias also present an untapped resource. Following President Volodymyr Zelenskyy’s example, these groups use social media to thank specific donors and upload videos of new weapons shipments. The Ukraine war has witnessed the emergence of fighter-influencers, whose account bios outline their position in the military and link to a PayPal account created for their audience to help fund the activities of their unit. In exchange, these fighters post regular updates of their activities on the frontlines. Individual fighters and civilians also use public Telegram channels where they post troves of information, spanning text, images, and videos, documenting how weapons are being deployed (and destroyed) on the ground.
The State Department should also introduce mechanisms for individual units, fighters, and civilians to directly report the presence of high-risk technologies in different areas of Ukraine. Using cellphone cameras, individuals on the ground could take photos of weapons, geocode the images to include locational data, and upload them to a central data repository to cross-reference with a list of weapons provided to Ukrainian partners.
Wherever possible, the State Department could introduce standards, such as requiring high-quality images of serial numbers and locational data in order to receive small monetary payments, and limits, such as a ceiling on the number of reported images an individual could upload within a defined period. While providing financial incentives may lead to false flags, manipulated imagery, or disinformation, verification would come with repeated reporting on the presence of a weapons technology by multiple sources. No singular data point would be sufficient on its own. Rather, several separate reports from multiple sources (and forms of data) would corroborate submissions as actionable intelligence. The program could also identify verified contributors, selected from an existing group with established credibility (ranging from journalists, civilians, and fighters), whose reporting could be granted weighted value.
By introducing payment incentives for civilian (and fighter) reporting, the U.S. could engage in cross-validation from multiple data sources. A final intelligence system for Ukraine’s end-use monitoring could include a combination of civilian reporting, satellite imagery, social media posts, and traditional, in-person review. Together, this data set would offer baseline information on the location and use of specific weapons in Ukraine.
Analyzing Large-Scale Multimodal Data on the Ukrainian Conflict
This proliferation of data will undoubtedly give rise to a new problem—sifting through it all. Supported by machine-learning tools, end-use monitors could parse through this rich data set to generate additional intelligence and better inform decision-making on military aid. This too has precedent—the U.S. need not develop training data from scratch. As a result of the private sector’s need for online content moderation of violence, many data sets exist to build and train artificial intelligence-based tools to recognize the presence and types of weapons in both video and images. Publicly available computer vision tools can sift through this growing online data set.
Many organizations, including an existing research base, have sought to fine-tune tools built for weapons recognition in both video and images. As an example, Google Lens, a publicly available image recognition technology, is able to discern and extract the serial number of the projectile in an image sourced from the aforementioned Twitter account (@UAWeapons, see Images 1 and 2). In this case, Google Lens extracted the writing on the projectile, identifying it as a M1712A1 artillery round used in howitzers. It also picked up on the smaller writing, stating that the projectile is from Lot A and the word “project.”
Dutch munitions posted by @UAWeapons.
Google Lens identifies the munition.
Google Lens suggests similar photos.
This type of tool could also be used to extract data from the written content of the associated post and to create preliminary data on the origin and current location of the projectile. In this tweet, the poster states that the projectile was likely provided by the Netherlands. This information could be further validated using other indicators in the image and other forms of intelligence.
While the State Department may already use OSINT for intelligence gathering within individual bureaus and offices, it remains unclear whether this use extends to nontraditional applications, such as end-use monitoring. Also uncertain is the degree to which these offices are making use of social media data, as compared to satellite imagery or conventional news reporting. This would be a missed opportunity. Given the private sector’s vocal and public support of the Ukrainian people, the U.S. government could more readily partner with companies for predeveloped, cutting-edge computer vision tools.
The Need for Innovative End-Use Monitoring
As the war in Ukraine rages on, demand for U.S. support will only continue. More recently, Ukraine has requested more weapons aid, including F-16 fighter jets. While the U.S. declined to provide this form of support, more requests will come. And importantly, the aid that’s already been provided requires proper and effective monitoring. For moral and morale reasons, the U.S. must prevent the proliferation of its weapons on the black market and Russian use of stolen Ukrainian weapons to commit war crimes or fabricate incidents. Investing in the creation of a baseline data set would enable the U.S. government to evaluate and modify its current monitoring plan and to anticipate vulnerabilities in the types of weapons most at risk. While there’s no replacement for on-the-ground, in-person monitoring, machine-learning-supported analysis of this rich social media archive stands to provide novel insights on the flow of weapons in Ukraine.