The New CBRN Paradigm: Accelerating Development on Both Sides

For over a century, the chemical, biological, radiological, and nuclear (CBRN) enterprise focused on chemicals, biological agents, radiological isotopes, and nuclear weapons proliferation. The shift from the Cold War to counterterrorism changed the employment of the mission, but the fundamentals have stayed the same. The science behind the ability of the United States to detect threats, protect Warfighters from those threats, decontaminate impact areas, and manage the consequences has advanced in a reasonably straight-line trajectory. A new threat may have been identified, or a new capability was conceptualized. A new detector was developed, such as the joint chemical agent detector, or a new decontaminant realized, such as reactive skin decontamination lotion. There are currently three emerging technologies and concepts that may enable and disrupt the CBRN enterprise, depending on who capitalizes on them first. They are as follows: (1) artificial intelligence (AI), (2) additive manufacturing (e.g., three-dimensional [3-D] printing), and (3) communications and networking of sensors.

AI

Although AI has become a buzzword, this does not detract from its importance. Leveraging AI will continue to change how the world does virtually everything. The CBRN enterprise is no different.

Recently, the U.S. Department of Homeland Security (DHS) Countering Weapons of Mass Destruction Office released the “Department of Homeland Security Report on Reducing the Risks at the Intersection of Artificial Intelligence and Chemical, Biological, Radiological, and Nuclear Threats” [1]. DHS identifies six findings on the misuse of AI to enable the development or production of CBRN threats. For each finding, several recommendations are made to enable the U.S. government to combat such threats.

Many AI tools and datasets are available in open-source locations. The DHS report speaks of AI democratization. Generally, “The most common goal of democratizing AI use is to distribute the benefits of AI use for many people to enjoy” [2]. There are several common open-source repositories for AI development. Rebecca Lipton, a data scientist who supports the Homeland Defense and Security Information Analysis Center (HDIAC) and partner Information Analysis Centers, explained, “GitLab and GitHub are two common code libraries/repositories where people/organizations publish code. For GitLab and GitHub, you can clone the code onto your machine, or oftentimes the author will bundle the code up into a package that can be downloaded neatly onto your machine using package/environment management software” [3]. This sounds quite easy for anyone with a small amount of software development knowledge to employ, which is the point of these sites. Yet, as Elizabeth Seger from the Carnegie Council for Ethics in International Affairs points out, “The circle of individuals who would greatly benefit from access to an AI drug discovery tool is relatively small (mainly pharmaceutical researchers); however, these tools can be repurposed to discover new toxins that might be used as chemical weapons” [2].

Another sharing site suggested by Lipton is called Hugging Face. This site houses a variety of open-source models, datasets, documents, and other solutions. The following example only took a few minutes to accomplish. By searching models simply for the word “chemical,” there were 90 returns. While some of the returns were incomplete results, a specific model, “biobert_chemical_ner,” populated. The developer provided a very simple explanation of the model [4]: “BioBERT model fine-tuned in NER task with BC5CDR-chemical and BC4CHEMD corpus,” which is defined and detailed next.

In translation, BioBERT is the Bidirectional Encoder Representations from Transformers for Biomedical Text Mining. Developed in 2019, it “is a domain-specific language representation model pre-trained on large-scale biomedical corpora” [5]. NER stands for named entity recognition, which is “the task of identifying and categorizing key information (entities) in text” [6]. BC5CDR is the BioCreative V chemical disease reaction corpus. The “BC5CDR corpus consists of 1,500 PubMed articles with 4,409 annotated chemicals, 5,818 diseases, and 3,116 chemical-disease interactions” [7]. BC4CHEMD is the BioCreative IV chemical compound and drug name recognition tool. “BC4CHEMD is a collection of 10,000 PubMed abstracts that contain a total of 84,355 chemical entity mentions labeled manually by expert chemistry literature curators” [8].

The simple “biobert_chemical_ner” model utilizes information in other models and databases to train and refine the BioBERT model to help classify and categorize massive volumes of biomedical information. Since this includes chemical disease reactions and interactions and tens of thousands of chemicals labeled by expert chemistry literature curators, considerable research has already been done for anyone wanting to use this tool. Whether a valid university researcher or someone with nefarious intentions, this information can be used to greatly reduce development time of biomedical cures and weapons alike. It should be noted that the “biobert_chemical_ner” model had over 9,000 downloads in the previous month—this is a large amount of white noise for bad actors to camouflage their intentions.

Naturally, AI can be employed to counter CBRN threats as well. The DHS report finds that “integration of AI into CBRN prevention, detection, response, and mitigation capabilities could yield important or emergent benefits” [1]. One of the recommendations tied to this finding is to “optimize the responsible use of AI in the design, testing, and evaluation of personal protective equipment, medical countermeasures (e.g., vaccines and synthetic antibodies), and decontaminants” [1]. The previously mentioned example of the “biobert_chemical_ner” model could be leveraged by developers to identify reactions to specific compounds to speed up development efforts.

For example, the Generative Unconstrained Intelligent Drug Engineering (GUIDE) program’s mission “is to leverage its integrated computational and experimental capabilities to accelerate drug development for the Warfighter by harnessing the power of advanced simulation and machine learning” [9]. This program, spearheaded by the Joint Program Executive Office (JPEO) for Chemical, Biological, Radiological, and Nuclear Defense (CBRND), seeks to adapt to the new reality of drug design leveraging AI and machine learning to mitigate the time and expense of investigating various drugs. Figure 1 demonstrates the GUIDE program’s user interface.

Additive Manufacturing

Additive manufacturing, especially 3-D printing, offers an even lower bar of entry than AI. 3-D printers come in all shapes and sizes and can be used to manufacture items out of multiple, different materials.

One limitation of employing CBRN munitions is the delivery system where nonstate actors have lacked access to methods to employ CBRN weapons in any meaningful way. In most cases, they would have to rely on getting close to the target and, in turn, endangering themselves. Even those who would be willing to commit suicide for their causes do not want to suffer the drawn-out death of a nerve agent, biological toxin, or radiation poisoning.

Among the most notorious terrorist attacks using chemical agents in history is the March 20, 1995, attack by the Aum Shinrikyo cult that killed at least 12 people and hospitalized nearly a thousand. Thousands more suffered effects from the sarin gas released in the subway. The sarin used was not high quality, thus reducing its effectiveness. Mercifully, the method to disperse the gas also reduced the effects. While traveling on trains in the Tokyo subway system, the gas was released from bags punctured by the assailants. This meant a gradual release from a single point on each train. While thousands were affected, the death toll would have been much higher had a better delivery method or chemical with better purity been used [10].

In bygone years, developing a delivery system for a weapon, conventional or CBRN, would require a large investment in machinery and expertise. That paradigm has shifted, and now anyone with access to a 3-D printer and basic materials can manufacture a delivery system.

In March of 2023, “the Relativity Space Terran 1 rocket lit up the night sky as it launched from Cape Canaveral Space Force Station in Florida. This was the first launch of a test rocket made entirely from 3-D-printed parts, measuring 100 feet tall and 7.5 feet wide” [11]. While this is an extreme example at the cutting edge of scientific research, the basic idea that a rocket engine can be produced using additive-manufacturing techniques proves the concept. Figure 2 shows the first additive-manufactured Glenn Research Copper (GRCop) combustion chamber.

In 2020, the Homeland Security Advisory Council released its “Final Report of the Emerging Technologies Subcommittee: 3D-Printing” [12]. Drawn from that report, Figure 3 forecasts 3-D-printing technology development.

Notably, the timelines in Figure 3 have already slid to the left, indicating novel methods of developing delivery systems and their CBRN threats.

Because of this, the United States is working hard to leverage 3-D-printing applications. There have been huge strides in manufacturing items such as replacement parts and medical devices for everyday military and civilian use. There are also successful use cases in laboratories and prototyping.

In the biomedical field, organ-on-a-chip (OOC) technology has allowed researchers to create small microchips that represent human organs and complex systems. The U.S. Army Combat Capabilities Development Command Chemical Biological Center recently completed testing using a 3-D-printed lung chip. According to their report [13]:

3-D printing technologies greatly simplify the traditional photolithography processes, reducing the need for experimental procedures and dramatically reducing processing costs and time. Integration between organ chip engineering and 3-D printing in the manufacturing process provides new opportunities for building more physiologically relevant organ chips in a more timely and cost-effective manner.

OOC is just one of many examples where 3-D printing will play a major role in reducing costs, procurement times, and design times. The potential for creating surgical devices, drugs, adaptive devices, and other items is only limited to the imagination. Although there are inherent safety concerns, such as the transmission of infectious diseases through a device that was not produced in a rigorously controlled setting, the future is promising.

Communications

Communications technology has exploded. For millennia, communications was restricted to the speed of the fastest horse. Now, worldwide, near-real-time communications are a reality. There are some restrictions that require their own solutions, such as bandwidth and security. Another major problem is that there is too much information for any one individual to manage.

Depending on the situation, there are a variety of solutions that may be available. Each one has benefits and risks, and so it is nearly impossible to find a solution, or even a suite of solutions, that solve all the problems.

One potential solution is edge computing. “Edge computing allows devices in remote locations to process data at the ‘edge’ of the network, either by the device or a local server. And when data needs to be processed in the central datacenter, only the most important data are transmitted, thereby minimizing latency” [14]. In other words, a CBRN detector may include, or be paired with, a method of communications. Instead of transmitting all the data it is receiving, it will only transmit the very basics needed to make a decision. As a hypothetical example, a detector designed to detect the presence of volatile organic compounds (VOCs) and oxygen levels may only send data if a certain level of either is reached. However, if that detector was retrieved, the timestamped data of oxygen levels and any trace VOCs or fluctuations in oxygen levels could be seen. A centralized system may also be set up to pull the data from the sensor on demand.

AI may be leveraged to assist the CBRN community with communications and maintaining situational awareness. Systems exist to translate voice to text (VTT). Conceptually, a VTT system could be trained and refined to understand terms and acronyms specific to the CBRN enterprise. This may include the transcription of specific radioisotopes. Cobalt-60 would be transcribed as how the commander wanted to read it. In a large event where multiple responders are communicating, another program could be scrubbing the text from the VTT and flagging any mentions of Cobalt-60 to the commander. Similarly, the VTT software could be trained to recognize certain chemicals, compounds, or other threats.

The U.S. military communicates on every possible band of the electromagnetic spectrum. While this is an enabler, it is also creating challenges when trying to network disparate sensors, unmanned systems, vehicles, aircraft, ships, etc. The answer for the CBRN community lies in the CBRN Support to Command-and-Control (CSC2) program. “CSC2 will provide integrated situational awareness about potential CBRN hazards to inform decision-making. CSC2 will link sensors together to develop networked tools that communicate and share information to achieve integration of CBRN capabilities and data with existing user systems across the service” [15]. Figure 4 shows the overview of CSC2 system architecture and its interoperability nodes.

This is a large undertaking. It is not as simple as networking the induvial sensors together. “The goal is for the networked sensors to be transmitted or ‘plugged into’ service-specific computing environments and ultimately, Joint All-Domain Command and Control (JADC2)-compatible CBRN Common Operating Environment” [15]. Currently, CSC2 may need to be interoperable with several different systems. The exact architecture still needs to be designed. While this problem is well known within the U.S. Department of Defense, the acceleration of technology means that the architecture becomes more complicated every day.

Still, CSC2 promises to be the CBRN community’s answer to communicating with the maneuver forces and combatant commanders to real-time decision-making. According to Paul Gietka, joint project lead for CBRN integration in the JPEO-CBRND, “CSC2 is our overarching system-to-system software capability that provides for the interoperability and integration of CBRN and non-CBRN sensors to achieve the needed situational awareness and understanding interdependent with service and mission partner computing environments” [17]. The program, at least at a high level, understands the challenges ahead.

There is one major drawback and perhaps a massive one—the increase in sensors and communications equipment increases the electromagnetic footprint that U.S. military forces will produce. When dealing with a peer or near-peer adversary in large-scale combat operations, this can quickly become a disadvantage. The problem is twofold: adversaries may be able to pick up the signal and thus target forces or equipment and adversaries could attempt to block signals, rendering the equipment useless. “One of the biggest lessons from Russia’s incursions in Ukraine—stemming from 2014 to its current invasion—is how units can be located and targeted with kinetic munitions solely based on their emissions within the electromagnetic spectrum” [18]. The need to sense CBRN threats in a tactical environment must be balanced with the potential for friendly forces being located simply by operating that equipment.

Conclusions

During the CBRN Defense Conference hosted by the National Defense Industrial Association, retired U.S. Army Brigadier General William King asserted that the CBRN enterprise is at an inflection point. This indicates a change in direction from the previous way of thinking and operating. Here, three major changes to how the CBRN community thinks and acts have been examined. AI, additive manufacturing, and communications are each a technology not specific to CBRN defense, but all will create radical changes in the CBRN landscape. On the flip side, adversaries can use these same technologies to accelerate their own capabilities. The inflection point truly is upon us.

Acknowledgments

Rebecca Lipton, a data scientist supporting HDIAC, provided simple answers to complicated questions to assist in the AI portion of this article. Her help was instrumental in the AI and Communications sections of this article.

References

1. DHS. “Department of Homeland Security Report on Reducing the Risks at the Intersection of Artificial Intelligence and Chemical, Biological, Radiological, and Nuclear Threats.” Homeland Security report, https://www.dhs.gov/sites/default/files/2024-06/24_0620_cwmd-dhs-cbrn-ai-eo-report-04262024-public-release.pdf, April 2024.
2. Seger, E. “What Do We Mean When We Talk About ‘AI Democratization?’” Carnegie Council for Ethics in International Affairs, https://www.carnegiecouncil.org/media/article/what-do-we-mean-when-we-talk-about-ai-democratization, 23 February 2024.
3. Lipton, R. Personal communication. SURVICE Engineering Company, Belcamp, MD, 27 June 2024.
4. Hugging Face. “Biobert_chemical_ner.” Hugging Face, https://huggingface.co/alvaroalon2/biobert_chemical_ner, accessed on 27 June 2024.
5. Lee, J., W. Yoon, S. Kim, D. Kim, S. Kim, C. Ho So, and J. Kang. “BioBERT: A Pre-Trained Biomedical Language Representation Model for Biomedical Text Mining.” Bioinformatics, pp. 1–7, https://arxiv.org/pdf/1901.08746, 2019.
6. Marshall, C. “What Is Named Entity Recognition (NER) and How Can I Use It?” Medium, https://medium.com/mysuperai/what-is-named-entity-recognition-ner-and-how-can-i-use-it-2b68cf6f545d, accessed on 27 June 2024.
7. Li, J., Y. Sun, R. J. Johnson, D. Sciaky, C. Wei, R. Leaman, A. P. Davis, C. J. Mattingly, T. C. Wiegers, and Z. Lu. “BioCreative V CDR Task Corpus: A Resource for Chemical Disease Relation Extraction.” Database: The Journal of Biological Databases and Curation, vol. 2016, https://doi.org/10.1093/database/baw068, accessed on 27 June 2024.
8. Papers With Code. “BC4CHEMD (BioCreative IV Chemical Compound and Drug Name Recognition).” Papers With Code, https://paperswithcode.com/dataset/bc4chemd, accessed on 27 June 2024.
9. Burkhalter, K. “GUIDE Program Prepares DOD to Defend Against the Unknown.” JPEO-CBRND: News, https://www.jpeocbrnd.osd.mil/Media/News/Article/3685461/guide-program-prepares-dod-to-defend-against-the-unknown/, 27 February 2024.
10. Olson, K. B. “Aum Shinrikyo: Once and Future Threat?” Emerging Infectious Diseases, vol. 5, issue 4, https://stacks.cdc.gov/view/cdc/14875/cdc_14875_DS1.pdf, July–August 1999.
11. Kilkenny, N. S. “3D Printed Rocket Launched Using Innovative NASA Alloy.” NASA: Glenn Communications, https://www.nasa.gov/centers-and-facilities/marshall/3d-printed-rocket-launched-using-innovative-nasa-alloy/, accessed on 27 June 2024.
12. DHS. “Final Report of the Emerging Technologies Subcommittee: 3D-Printing.” Homeland Security Advisory Council, https://www.dhs.gov/sites/default/files/publications/final_report_hsac_emerging_technology_subcommittee_3dprinting_508_compliant.pdf, 24 February 2020.
13. Fudge, D. H., P. Lee, R. Evans, T. D. P. Goralski, and B. Ruprecht. “Design and Application of an Adept Aerosol Lung-on-a-Chip and Aerosol and Vapor Delivery Systems.” DEVCOM CBC-TR-1888, U.S. Army Combat Capabilities Development Command Chemical Biological Center, Aberdeen Proving Ground, MD, June 2024.
14. Microsoft. “What Is Edge Computing?” Microsoft, https://azure.microsoft.com/en-us/resources/cloud-computing-dictionary/what-is-edge-computing, accessed on 27 June 2024.
15. Burkhalter, K. “CBRN Defense, Computer Software, and Command Decisions: CSC2 Program Supports the Joint Force through Integrated Software Tools.” JPEO-CBRND: News, https://www.jpeocbrnd.osd.mil/Media/News/Article/3568671/cbrn-defense-computer-software-and-command-decisions-csc2-program-supports-the/, 25 October 2023.
16. Murphy, D. “CBRN Support to Command & Control (CSC2).” Joint Product Leader for CBRN Support to Command and Control intelink presentation, https://intelshare.intelink.gov/SitePages/Home.aspx, 7 March 2022.
17. Carberry, S. “JUST IN: New Sensors, Comms Needed for CBRN Threat.” National Defense: NDIA’s Business & Technology Magazine, https://www.nationaldefensemagazine.org/articles/2024/6/25/new-sensors-comms-needed-for-cbrn-threat, 25 June 2024.
18. Pomerleau, M. “Army Wants to ‘Reimagine’ Electromagnetic Signature Management.” Defensescoop, https://defensescoop.com/2023/10/11/army-wants-to-reimagine-electromagnetic-signature-management/, 11 October 2023.

Biography

John Clements is the technical lead for HDIAC. He served 20 years as a combat engineer in the U.S. Marine Corps Reserve and deployed three times to Iraq in support of Operation Iraqi Freedom. His prior work includes test and evaluation on procedures and systems related to CBRN decontamination, mortuary affairs, cyber insider threat, open-source and social media information, the Common Operational Picture used by combatant commands, and the Mounted Computing Environment. He has extensive experience working with joint, interagency, and allied partners at the strategic and tactical levels. Mr. Clements holds an M.A. in homeland security from the American Military University.