Closed hearings to examine harnessing artificial intelligence cyber capabilities; to be immediately followed by an open hearing at 3:30 p.m. in SR-232A.
Meeting•Senate Armed Services Subcommittee on Cybersecurity•Mar 25, 2025 · 2:30 PM
Summary
Senate Armed Services Subcommittee on Cybersecurity held a meeting on Mar 25, 2025 at 2:30 PM in Capitol Visitor Center (Senate side, Room 217.
Record
The meeting has its transcript on the record.
Transcript
The transcript runs to 1,206 lines and 63,616 characters, as the Government Publishing Office printed it.
1[Senate Hearing 119-107]2[From the U.S. Government Publishing Office]34 S. Hrg. 119-10756 HARNESSING ARTIFICIAL INTELLIGENCE CYBER7 CAPABILITIES89=======================================================================1011 HEARING1213 BEFORE THE1415 SUBCOMMITTEE ON16 CYBERSECURITY1718 OF THE1920 COMMITTEE ON ARMED SERVICES21 UNITED STATES SENATE2223 ONE HUNDRED NINETEENTH CONGRESS2425 FIRST SESSION2627 __________2829 MARCH 25, 20253031 __________3233 Printed for the use of the Committee on Armed Services3435[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]3637 Available via: http://www.govinfo.gov3839 __________4041 U.S. GOVERNMENT PUBLISHING OFFICE4260-836 PDF WASHINGTON : 20254344-----------------------------------------------------------------------------------4546 COMMITTEE ON ARMED SERVICES4748 ROGER F. WICKER, Mississippi, Chairman4950DEB FISCHER, Nebraska JACK REED, Rhode Island51TOM COTTON, Arkansas JEANNE SHAHEEN, New Hampshire52MIKE ROUNDS, South Dakota KIRSTEN E. GILLIBRAND, New York53JONI ERNST, Iowa RICHARD BLUMENTHAL, Connecticut54DAN SULLIVAN, Alaska MAZIE K. HIRONO, Hawaii55KEVIN CRAMER, North Dakota TIM KAINE, Virginia56RICK SCOTT, Florida ANGUS S. KING, Jr., Maine57TOMMY TUBERVILLE, Alabama ELIZABETH WARREN, Massachusetts58MARKWAYNE MULLIN, Oklahoma GARY C. PETERS, Michigan59TED BUDD, North Carolina TAMMY DUCKWORTH, Illinois60ERIC SCHMITT, Missouri JACKY ROSEN, Nevada61JIM BANKS, INDIANA MARK KELLY, Arizona62TIM SHEEHY, MONTANA ELISSA SLOTKIN, MICHIGAN6364 John P. Keast, Staff Director65 Elizabeth L. King, Minority Staff Director6667_________________________________________________________________6869 Subcommittee on Cybersecurity7071 MIKE ROUNDS, South Dakota,72 Chairman73TOM COTTON, Arkansas JACKY ROSEN, Nevada74JONI K. ERNST, Iowa KIRSTEN E. GILLIBRAND, New York75TED BUDD, North Carolina GARY C. PETERS, Michigan76ERIC SCHMITT, Missouri ELISSA SLOTKIN, Michigan7778 (ii)7980 C O N T E N T S8182_________________________________________________________________8384 march 25, 20258586 Page8788Harnessing Artificial Intelligence Cyber Capabilities............ 18990 Members Statements9192Statement of Senator Mike Rounds................................. 19394Statement of Senator Jacky Rosen................................. 29596 Witness Statements9798Mitre, Mr. Jim, Vice President and Director, Rand Global and 399 Emerging Risks.100101Tadross, Mr. Dan, Head of Public Sector, Scale AI................ 10102103Ferris, Mr. David, Global Head of Public Sector, Cohere.......... 19104105 (iii)106107 HARNESSING ARTIFICIAL INTELLIGENCE CYBER CAPABILITIES108109 ----------110111 TUESDAY, MARCH 25, 2025112113 United States Senate,114 Subcommittee on Cybersecurity,115 Committee on Armed Services,116 Washington, DC.117 The Committee met, pursuant to notice, at 3:31 p.m. in room118SR-232A, Russell Senate Office Building, Senator Mike Rounds119(Chairman of the Subcommittee) presiding.120 Committee Members present: Senators Rounds, and Rosen.121122 OPENING STATEMENT OF SENATOR MIKE ROUNDS123124 Senator Rounds. Good afternoon, and I'd like to thank our125witnesses for appearing today to discuss how artificial126intelligence can be utilized to enhance the Department of127Defense's (DOD) cyber capabilities. We have just heard from128experts in our closed session from the U.S. Cyber Command, the129Defense Advanced Research Projects Agency (DARPA), and the130DOD's Chief Digital and Artificial Intelligence Office. These131organizations all play a crucial role in making sure the132Department is postured to carry out its national security133mission in cyber space.134 Recent cyberattacks against U.S. critical infrastructure135are a stark reminder of the growing sophistication and136persistence of cyber threat actors. To outpace our adversaries137in the cyber domain, the Department must rapidly harness the138advances of AI [Artificial Intelligence] technologies. This139means that the Department of Defense needs capable partners140outside of the Pentagon who are moving at breakneck speed to141solve our national security challenges.142 This brings us to our hearing topic today; how the143Department can leverage AI-enabled capabilities to field144exquisite, offensive, and defensive cyber tools, enhance our145ability to detect cyber threats, and automate threat mitigation146to gain an enduring advantage in cyberspace.147 I also look forward to hearing from the witnesses about how148the Department can be better equipped to counter enemy AI-149enabled cyber capabilities, and leverage AI to enhance our150overall war fighting ability in the cyber domain. Our151innovators and tech companies are one of our asymmetric152advantages in the cyber fight, but the gap is steadily closing.153 At the tip of the spear is artificial intelligence.154Unfortunately, the Chinese Communist Party understands this all155too well. Xi Jinping has spoken about the importance of AI.156With the release of DeepSeek earlier this year, it is clear157unless we act decisively and soon, China will not be playing158catch up. We will.159 U.S. advancements in this critical technology are160impressive, and we are fortunate to have some of the best161innovators in the world. As Silicon Valley and other leading162technology developers continue their research and development163of AI at the bleeding edge, our job must be to integrate those164tools in a secure, but rapid fashion into our cyber165capabilities.166 I look forward to hearing from our witnesses who all bring167unique and firsthand experience about how the Department can168speed up its use of AI in the cyber domain. Again, thank you to169our witnesses for coming here today.170 Before I introduce them, I'll now recognize Ranking Member171Senator Rosen.172173 STATEMENT OF SENATOR JACKY ROSEN174175 Senator Rosen. Well, thank you, Chairman Rounds, and I'd176like to begin by welcoming our panel, and thanking you all for177joining us. This topic has profound implications for our178national security, I would say, for our personal security, for179everything in our world to come.180 But this is actually my first hearing as Ranking Member of181this Subcommittee, and I am really honored to work alongside182Chairman Rounds, our colleagues, and each of you on how we can183responsibly integrate innovation and the increasing pace of184technology including artificial intelligence into our national185defense strategy and into the hands of our service members to186enhance their speed, their capabilities, and their operating187picture. Well, of course, all the time we have to balance the188risks and rewards concerns of AI and what it teaches us.189 So, with great promise comes great responsibility. We know190that our adversaries are developing new AI tools and have the191potential to fundamentally shift the nature of warfare. We've192began to see how new uses of AI can help our own service193members counter such threats and take proactive offensive194actions in the moment as well.195 However, the rapid pace of AI innovation also raises really196important questions about its ethical implications, its197governance, and the security risks it poses as well. We're198operating in a new world without guardrails and we need to199tread carefully, balancing such caution with the need to create200an environment that allows for innovation and agility.201 There are also challenges we must overcome in order to both202mitigate the risks of AI and make the most of the opportunities203that I know it presents. In particular, we need to further204invest in and expand the AI workforce, both at DOD, and across205the Government, across the private sector. We have to increase206it everywhere to harness our full potential. I truly believe207this.208 As a former computer programmer, systems analyst, myself, I209can say from firsthand experience that AI has vastly changed210the technology landscape since I began my career. Many of the211coding and the programming skills that people like me brought212to the table, which form the backbone of what CYBERCOM213personnel do every day, in both offensive and defensive214operations, can now be supplemented by AI.215 I know it doesn't replace us, that's for sure. But however,216this does pose its own set of risks. It creates a deep need for217us to invest in that new kind of cyber workforce that is218centered around understanding these AI skills, and we continue219to have a cyber and AI skills gap.220 Until we meet that challenge of bridging it, understanding221it, being able to see its potential, and at the same time222understand how it improves our own potential as human beings,223we're going to continue to be at the risk of our adversaries224having the upper hand.225 So, I look forward to discussing such challenges today and226over the course of this Congress. I thank our panel once again227for your expertise and contributions to that effort, and I228thank you again, Mr. Chairman.229 Senator Rounds. Thank you, and it is a pleasure to have you230here on the team with us. This is one of those subcommittees in231which it is very bipartisan, and we have focused on this since232the creation of this by Senator McCain back in 2017, I believe.233The path forward, I think, has been made better because of the234work that we've done in the past on a bipartisan basis to keep235everything on the straight and narrow.236 I want to thank all of you once again for coming in and237participating in this open session, and we have with us, today,238all three of you here. Beginning with Mr. Jim Mitre, Vice239President and Director of RAND Global and Emerging risks. Mr.240Mitre, welcome. Mr. David Ferris, Global Head of Public Sector,241Cohere. Welcome, and Mr. Dan Tadros, Head of Public Sector,242Scale AI.243 I understand that the agreement has been made that Mr.244Mitre, you will begin today. So, we welcome you for your245opening statement, sir.246247 STATEMENT OF MR. JIM MITRE, VICE PRESIDENT AND DIRECTOR, RAND248 GLOBAL AND EMERGING RISKS249250 Mr. Mitre. Terrific. Chairman Rounds, Ranking Member Rosen,251thank you so much for the opportunity to testify today on the252national security implications posed by the potential emergence253of advanced artificial intelligence, or artificial general254intelligence, AGI.255 Leading AI companies in the United States, China, and the256rest of the world, are in hot pursuit of AGI, which would257possess human level or potentially even superhuman level258intelligence across a wide variety of cognitive tasks. The pace259and potential progress of AGI's emergence, as well as the260composition of a post-AGI future, are uncertain and hotly261debated. Yet the emergence of AGI is plausible and the262consequences so profound that the U.S. national security263community should take it seriously and plan for it.264 Consider the following. What would the U.S. Government do265if in the next few years, a leading AI company announced that266its forthcoming model had the ability to produce the equivalent267of 1 million computer programmers as capable as the top 1268percent of human programmers at the touch of a button. The269national security implications are substantial and could cause270a significant disruption of the current cyber offense defense271balance.272 At RAND, we are planning for it. Our work has revealed that273AGI presents five hard national security problems. First, AGI274might enable a significant first-mover advantage via the sudden275emergence of a decisive wonder weapon. For example, a276capability so proficient at identifying and exploiting277vulnerabilities in enemy cyber defenses, that it provides what278might be called a splendid first cyber strike, that completely279disables a retaliatory cyber strike. Such a first mover280advantage could disrupt the military balance of power in key281theaters, create a host of proliferation risks, and accelerate282technological race dynamics.283 Second, AGI might cause a systemic shift in the instruments284of national power that alters the balance of global power. The285history of military innovation suggests that being able to286adopt a new technology is more consequential than being the287first to achieve a specific scientific or technological288breakthrough.289 As the U.S. allied and rival militaries establish access to290AGI and adopted it at scale, it could upend military balances291by affecting key building blocks of military competition such292as hiders versus finders, precision versus mass, or centralized293versus decentralized command and control. States that are294better postured to capitalize on and manage systemic shifts295caused by AGI could have greatly expanded influence.296 Third, AGI might serve as a malicious mentor that explains297and contextualizes the specific steps that non-experts can take298to develop dangerous weapons such as violent cyber malware,299widening the pool of people capable of creating such threats.300 Fourth, AGI might achieve enough autonomy and behave with301enough agency to be considered an independent actor on the302global stage. Consider an AGI with advanced computer303programming abilities that is able to break out of the box and304engage with the world across cyberspace. It could possess305agency beyond human control, operate autonomously, and make306decisions with far reaching consequences.307 Fifth, the pursuit of AGI could foster a period of308instability as nations and corporations race to achieve309dominance in this transformative technology. This competition310might lead to heightened tensions reminiscent of the nuclear311arms race, such that the quest for superiority risks triggering312rather than deterring conflict. Misinterpretations or313miscalculations could precipitate preemptive strategies or arms314buildups that destabilize global security.315 As the U.S. Department of Defense embarks on developing the316National Defense Strategy, it will have to grapple with how317advanced AI will affect cyber along with all other domains. The318five hard problems that AGI presents to national security can319serve as a rubric to evaluate how the strategy addresses the320potential emergence of AGI.321 Thank you for the opportunity to testify. I welcome your322questions.323 [The prepared statement of Mr. Jim Mitre follows:]324325 [GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]326327 Senator Rounds. I thank you. Mr. Tadross, unless you folks328have agreed on a different. Mr. Tadross.329330 STATEMENT OF MR. DAN TADROSS, HEAD OF PUBLIC SECTOR, SCALE AI331332 Mr. Tadross. Chairman Rounds, Ranking Member Rosen, Members333of the Subcommittee, thank you for the opportunity to be here334today.335 My name is Dan Tadross. I lead Scale AI's public sector336business. Every day, my team is singularly focused on how to337bring best-in-class AI into the DOD and other agencies. Scale338was founded in 2016, and since that time, has powered nearly339every AI innovation. Our role in this critical ecosystem340provides us a unique opportunity to understand how to build341high quality AI systems powered by the world's best data.342 Our work is deeply personal to me as I have worked nearly343my entire career at the intersection of AI and the Government.344During my time as an Active Duty marine, I had the privilege of345helping to stand up the Joint Artificial Intelligence Center,346which enabled me to see firsthand the challenges and struggles347associated with the DOD's implementation of AI.348 This hearing comes at a critical time for the future of AI349leadership, and before we discuss what the United States must350do to win, it's important to analyze where things stand today.351 AI is made up of three main pillars; compute, data, and352algorithms. More than 1 year ago, the United States was clearly353ahead on all three. However, today, that is no longer the case.354Advancements from China have shown that they've closed the gap.355Today, China is leading on data. We're tied on algorithms, but356the United States remains ahead on compute. It's clear that the357race is neck and neck.358 In order to compete more aggressively, the CCP [Chinese359Communist Party] has implemented a whole-of-country approach to360accelerating its pursuit of becoming a global standard for AI361from an investment standpoint. For the first time in history,362China is benchmarking AI investment off the leading tech363companies and not the United States Government.364 Last year, China spent at least $1.2 billion on data365labeling alone compared to our under $100 million by the United366States. As part of China's AI Plus initiative, the Government367established seven data labeling centers around the country to368mainly support public sector application.369 Beyond data, while the U.S. has been stuck in a research370and pilot mindset, the CCP has rapidly increased their371investment in fielding AI capabilities. In the first half of3722024 alone, the PLA [People's Liberation Party] issued 81373contracts with large language model companies to rapidly grow374their capability. To win, the U.S. needs to unleash our375technology to the warfighter at an unprecedented pace.376 When it comes to adopting and implementing AI, the DOD has377not launched a new AI program in nearly a decade. For the past3784 years, DOD leadership spent countless hours developing379potential use cases for AI, researching and piloting AI380systems, and even putting out guidance to stop users from381utilizing AI.382 We still have time, but the window is closing. If we want383to win, we must not only buy into a vision, but it also takes384three clear and decisive actions. Number one, is put the right385AI foundation in place. To start, the DOD lacks the foundation386piece, the foundational pieces necessary to build, scale, and387implement widespread AI solutions. This needs to change, and we388must put in place the elements necessary to expand the use of389AI programs, and this starts with data.390 To truly prioritize and execute the strategy, it requires391two main aspects; AI-ready data requirements, and enterprise-392wide AI data infrastructure. The U.S. Government is the world's393leading producer of both quantity and diverseness of data. But394nearly all that data is going unused. If the U.S. wants to turn395our data into an advantage, this must change.396 In multiple NDAAs [National Defense Authorization Acts],397his committee has directed, suggested, and tried to require the398DOD to prioritize AI-ready data requirements, but it's clear399that more must be done. In parallel to implementing the400requirement, the Department should also set up enterprise-wide401AI data infrastructure.402 This commercial best practice ensures that AI programs are403developed in the most efficient and cost-effective manner, and404leading tech companies have long realized this requirement for405effectiveness. For that reason, China is mirroring this same406approach.407 Number two, is to shift our mindset to be an408implementation-first. If the U.S. is going to win, we must409shift into an implementation-first mindset. In order for this410to occur, Scale believes that the DOD must set must first set a411North Star related to robust AI implementation in no more than4125 years.413 This should focus on agentic applications such as agentic414warfare, and would provide an ambitious vision and enable415tangible multi-year plan to reach it. Scale is actively working416on deploying the first instance of this in INDOPACOM [United417States Indo-Pacific Command] and EUCOM [United States European418Command] through DIU's [Defense Innovation Unit] Thunderforge419effort.420 Number three, is to ensure our acquisition system no longer421slows us down. AI is unique in that it is software, but needs422to be maintained like hardware, which presents challenges for423the DOD given that it doesn't neatly fit into a legacy424acquisition system. Congress took a strong first step by425requiring the DOD to break out AI elements of programs in the426future budgets, and it is critical that Congress continues to427provide oversight to push the DOD to do so quickly as possible.428 In addition to proposals like the FoRGED Act, Scale also429believes that we need to continue to look at finding ways to430break through the challenges of multi-year budgeting, which is431clearly still holding back the DOD's implementation of AI. With432these three decisive actions, the DOD will be better positioned433to adopt and effectively implement AI solutions.434 Thank you again for the opportunity to be here, and I look435forward to your questions.436 [The prepared statement of Mr. Dan Tadross follows:]437438 [GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]439440 Senator Rounds. Thank you very much, sir. Mr. Ferris.441442 STATEMENT OF MR. DAVID FERRIS, GLOBAL HEAD OF PUBLIC SECTOR,443 COHERE444445 Mr. Ferris. Chairman Rounds, Ranking Member Rosen,446distinguished Members of the Subcommittee, thank you for the447opportunity to testify today.448 My name is Dave Ferris, and I'm the Head of Global Public449Sector at Cohere. I previously served nearly 17 years in the450Canadian Armed Forces, including deployments to Afghanistan and451Ukraine, and spent the last 2 years of my career on the U.S.452Joint Staff in the Pentagon.453 Cohere is a leader in building AI systems designed454exclusively for government and enterprise use, prioritizing455privacy, security, multilingual capability, and verifiability.456Our expertise spans from building foundational AI models, to457developing AgentX systems. We focus on operationalizing AI,458integrating it into real missions, under real world459constraints. We partner with allied governments, agencies, and460leading global companies.461 Our primary goal is seamless integration, deep462customization, and accessible solutions that deliver immediate463practical value and confidence. We specialize in private464deployments, even air gapped environments where we do not see465our customer's data.466 Today, I would like to highlight four key topics of focus467gleaned from having worked with high security cyber defense468government organizations. The first key topic is how AI can469significantly enhance the Department of Defense's mission,470particularly in cybersecurity and intelligence.471 AI systems can dramatically improve pattern recognition and472anomaly detection across vast data sets. They can be invaluable473for sorting through and synthesizing huge volumes of multi-474source information, and they can help automate a number of475crucial tasks to provide early warnings and free humans to476focus on making strategic decisions.477 Similarly, effective AI adoption requires integrating478technology thoughtfully with existing workflows. Human AI479teaming is crucial in ensuring AI tools have user-friendly480interfaces. It helps build trust and maximizes operational481value.482 A second key topic is to consider how AI can help fight483back against competitor nations and malicious actors that are484already employing AI-enabled cyber capabilities. Reports have485shown these countries are automating their intrusion attempts486using AI to generate deceptive deep fakes, develop more487convincing phishing lures, and create information warfare.488 To stay ahead of these AI augmented threats, DOD must489likewise incorporate AI across its offensive and defensive490cyber operations. Large language models provide a unique491ability beyond traditional, rule-based machine learning systems492for language understanding and reasoning capabilities that493allows for dynamic identification, analysis, and generation of494conclusions across a wide range of use cases.495 The third key topic is to understand how technical496considerations are critical to successful AI deployments in497defense. Models should be right-sized for their specific498mission. Specialized efficient AI models can often outperform499larger general-purpose systems. This enables deployment even on500limited hardware such as edge devices like laptops or501classified data centers.502 Flexible secure deployment architecture is critical. AI503systems must be deployable across multiple secure environments504and ensure AI sovereignty. Similarly, ensuring models are505hardware agnostic and interoperable, so there is no lock into506one cloud or one chip provider, is essential to ensuring supply507chain and operational security.508 Collaborative development through public-private509partnerships allows for rapid customization of or creation of510new AI models to meet specific operational context while511protecting sensitive information. The DOD does not need to512undertake the costly, time-consuming task of developing every513AI model from scratch.514 The final key point is to highlight that Congress can take515immediate action to accelerate responsible AI adoption.516Congress should modernize procurement processes to allow517innovative AI startups easier entry. Procurement should reward518innovation, agility, and performance, not just size or past519contracts. New legislation should promote interoperability, and520open standards to prevent vendor-locking and enable diverse AI521solutions to seamlessly integrate into defense ecosystems.522 Finally, Congress should support robust internal523benchmarking and testing specific defense applications rather524than the use of generic academic benchmarks. This would ensure525AI reliability and trustworthiness in critical missions.526 In conclusion, Cohere is committed to partnering with DOD527in Congress ensuring AI tools are secure, effective, and528mission-ready. Thank you, and I look forward to your questions.529 [The prepared statement of Mr. David Ferris follows:]530531 [GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]532533 Senator Rounds. First of all, thank you to all of you, and534I appreciated your opening comments. We'll pass this back and535forth a little bit with regard to questions and so forth, but536we'll try to get to as many as we can in a short period of537time.538 I wanted to begin, Mr. Mitre. The artificial intelligence539is here to stay. It's not going away. You gave us some warning540signs out there, but I wanted to hear from you. We can't slow541down on the development of AI, or we know that our competitors542will clearly outpace us.543 Give me your rendition of how we do this without losing544facts or losing sight of the facts that there can also be some545dangers involved. You've identified a number of the possible546dangers, but how are we going to do this and still keep that in547mind?548 Mr. Mitre. That's a great question, and I welcome it. I549wholeheartedly agree that it's in America's interest to stay at550the forefront of the development of generative AI and AI551technologies more broadly.552 So, the way in which we can address this issue is, first,553it's helpful for the U.S. Government to really understand what554the current State of the technology is, and make sure that555folks within the Government, particularly those that are556working in the national security community, really understand557what's happening with the technology.558 Because one of the challenges with this technology is that559it's not being developed by Government, it's being developed by560the private sector. So, just understanding what the current561State is critical so there aren't technological surprises that562come out that shock people in the national security community.563 The second thing that Government should be doing here is564really looking for applications in the national security565context. What are the specific use cases that it can be566applied? What are potential pathways to wonder weapon or ways567in which it could be highly advantageous in a military568competition that's critical to do, and that means having the AI569in an environment where you've got sufficient compute, where570you've got the right networks, et cetera. You can actively571experiment with it, and get the technology in the hands of the572operators to play around with it.573 The third thing is preparing for contingencies. There's a574wide range of possible things that could happen. A loss of575control scenario, for example, areas where there is576technological surprise and the Chinese get ahead. What would577the U.S. Government do in such contingencies? We should think578that through in advance and have plans ready to address it.579 Senator Rounds. Thank you. Mr. Tadross, this works right580into some of the comments that you had made, and I want to581just, number one, I think it would be a statement we would all582agree on that continuing resolutions are absolutely not the583long-term plan that we need.584 If we're going to be able to move forward with the585investment in AI that we need, that may very well save a lot of586lives in the battlefield. So, I would recognize that up front,587and I think you were rather suggesting that a little bit in588terms of our failure to keep up with the demands of how quickly589AI is developing elsewhere.590 You also said something else, though, and I wanted to touch591on two items. Number one, you talked about the fact that we592have data, which is unused. I want you to explain that a little593bit, and then, second, of all feeding into to what Mr. Mitre594talked about, you talked about agentic warfare.595 Can you talk a little bit about what that really means for596the--I mean, we've got a lot of folks out here that this may be597their first introduction to the coordination of different598applications that are directly involved in warfare versus the599application of AI in general. So, first of all, data unused,600and second of all, agentic warfare.601 Mr. Tadross. Of course, Senator, and thank you for the602question. So, in terms of data being unused, the approach that603I was kind of looking at there is the aspect that, right, now604an enormous amount of information is being collected day to605day. But to take kind of a quote from one of the previous606Secretaries of the Air Force, ``We treat data like exhaust as607opposed to something that's really critical to use.''608 So, as a result, every time that we run an exercise, run a609command post exercise in terms of large amounts of chat data is610being developed, large amounts of chat data is being traced611back and forth, what's happening is at the end of that612exercise, all of those hard drives are just being purged or613being neglected and goes into storage.614 So, those are instances where the interactions between615participants of a staff, for example, should be getting616captured, and we should be using that to help develop training617data to using it to help develop benchmarks against how these618algorithms should operate. Then by doing so, are eventual619development of agentic solutions can be more in line with what620is required by those end users, which I think then brings us621into the idea of like agentic warfare.622 Really what that means, my interpretation of this, is we're623trying to move humans, move to a position from humans are the624loop to humans on the loop. So, right now, if a staff at625INDOPACOM, or at EUCOM, or any other combatant command needs to626make a decision, the process at which they do that hasn't627really changed since the advent of the Napoleonic staff628structure. We take the problem, we divide it up, and then629what's required is that the commander at the last minute has to630synthesize all of those things together and then make an631informed decision.632 The effort of agentic warfare is to move to the point where633much of that low-level staff work can be done by these AI634agents through automated methods with human oversight and635supervision of the process. It's important to maintain some636human oversight of the entire process to ensure that human-637context judgment, and the competitive advantage of the U.S.638military, which is the fact that we have the most well-trained,639well-versed staff and NCOs on the globe.640 Senator Rounds. Thank you. Mr. Ferris, I've got some641questions for you as well, but my first 5 minutes is up. We642will do a second round, but at this point, I'll come back to643Senator Rosen.644 Senator Rosen. Thank you. You know, I want to talk a little645about guardrails and benchmarks. Both, I believe they go hand646in hand. Over the last year, discussions between Congress,647prior administrations, they've always centered around trying to648come up with guardrails to promote responsible AI. You all know649what I'm talking about; nobody wants it to become an unchecked650technology.651 The current administration has raised concerns that652guardrails might inhibit innovation. I believe we need both653effective guardrails and benchmarks because the benchmarks,654just as if your child goes to school, they're the test to show655if they're learning and going in the direction that you're656expecting them to go. That's what's going to keep that circle657in check.658 So, I'm going to have questions for all three of you, but659I'll start they're similar, but I'm going to start with you,660Mr. Mitre. How should we develop guidelines, or the guardrails,661and benchmarks in ways that mitigate risk without stifling662innovation?663 I might also add, I'm actually going to ask all three of664you this. How do we develop, for those of us sitting in this665seat with all of you, a common policy language that is both666nimble, but provides the availability for us to do effective667oversight?668 Mr. Mitre. Thank you, Senator. So, I wholeheartedly agree669that it's important for us to understand what these models are670capable of doing, right? They're developed, and they're671released into the world with no user manual. It's not entirely672clear what applications they'll be able to perform or how673capable they'll be at doing that.674 So, benchmarks are crucial, particularly in a national675security context. It's helpful to understand what might the676latest generation model be able to do in terms of offensive677cyber defensive, cyber capabilities in terms of potentially678informing non-experts on how they go about designing a679bioweapon that could be highly transmissible and lethal, et680cetera. So, the real focus that is warranted is on developing681benchmarks to really just evaluate and understand what the682risks are.683 Separate question in terms of what should Government do684about those risks if they emerge, and should regulations or685something along those lines be appropriate in that regard? I686defer to Government for specific thoughts on that. What we're687trying to do is just understand at first pass what are some of688the risks here and make sure that people are well informed on689that point.690 Senator Rosen. Thank you, and I'm going to just go down.691Mr. Tadross, the same thing. Developing the guardrails. The692benchmarks tell us one thing, the guardrails tell us another. I693guess I'll make it all the same question. We are going to694struggle. We have to put this down in some way on paper that695allows us to be nimble and provide that ability to do the696oversight we need to.697 So, if you have thoughts about how we develop this common698language that we can all speak from or start from, I think is699really critical, so.700 Mr. Tadross. Absolutely. So, the way that our company kind701of looks at this, at least as it relates to guardrails in the702implementation of AI in the Department of Defense, is to really703look at it from a perspective of people, process, and704technology. That while the technology needs to have guardrails705by itself in terms of like its responses when it will trigger a706refusal, or when it may not, there still needs to be the other707two portions of this triangle.708 So, people need to be trained on how to best leverage the709capability. Then, the process needs to be adapted. Because if710we just bolt AI onto an existing process, then the advantages711are somewhat lost. So, the doctrine and training of the712individuals needs to adapt at the same time as the technology713has fielded.714 This goes back to my position about implementation. The715only way to do this is to experiment in low-risk environments716and to iterate very quickly. short of that, I'm afraid that the717concern about trying to write out the full answer at the718beginning of the test is probably unlikely. So, you need to be719able to learn from doing and be able to build off of that.720 As it relates to benchmarks, this is an area where our721company's done quite a bit of interesting work. So, we have a722paper that we've published showing that most of these large723language models and AI systems will essentially cheat off of724existing benchmarks. They've seen them, they understand the725rules of the test, and as a result, they will score abnormally726high.727 The approach that we've taken in partnership with728organizations like CSIS [Center for Strategic and International729Studies] and the CDAO [Chief Digital and Artificial730Intelligence Office] is to build custom benchmarks that are731focused on the domain at which it actually matters to test. So,732we've built these custom benchmarks. The algorithms have never733seen them, they've never been incorporated in their training734data. As a result, you can have a little bit more faith in the735performance of those algorithms.736 Senator Rosen. Thank you. Mr. Ferris?737 Mr. Ferris. Thank you, Senator. I echo the sentiment of my738colleague on the panel here. I think public benchmarks can739often be gamed. I'll start from the perspective of benchmarks740because I think it's relevant to what my colleague was saying.741They don't typically show the performance in real-world742context. So, we would----743 Senator Rosen. Is using the word ``audit'' better than744benchmark?745 Mr. Ferris. Well, no, I think we would say creating custom746benchmarks.747 Senator Rosen. Just like right-sizing your model.748 Mr. Ferris. Yes, exactly. Okay, and, you know, to kind of749take that down one step further, we work very closely with our750customers from beginning to end in order to ensure that we're751right-sizing that model, developing the benchmarks. But that752also includes some human evaluations because that human AI753interface is obviously imperative as we're moving down this.754 With respect to guardrails, you know, there's this healthy755tension between accountability and agility, I would say, in756this environment. So right now, we obviously would suggest that757we want to lean into the agility. We want to take an adoption758mindset, but can't, you know, sacrifice really the security759reliability and verifiability.760 So, you know, ensuring that you have clear visualization761into the data lineage, ensuring that you have a good762understanding of how those safety measures have been built into763the model during its development and deployment, I think, is764imperative.765 Senator Rosen. Well, I think because you say you want to766lean in to--oops, I'm going over my time. I'm sorry. Can I767finish the thought? Lean into the agility, but if you don't768keep humans, if you don't keep someone else in the loop,769people's lives are on the line. It's still a computer just770analyzing data, and so, at that execution point, you have to771consider leaning into agility. But at what execution points do772we allow for a better decision? I'll let it go to my--maybe773that's a philosophical question.774 Senator Rounds. Well, look here, and I'm going to lead into775this a little bit, too. I'm going to start with Mr. Ferris. We776talked about right-sizing systems, and kind of along the same777line here, I'm going to compare that because I'm not sure if778I'm thinking the same thing that you're proposing.779 But loitering, munitions as an example, we have clear780evidence that in the Nagorno-Karabakh War between Azerbaijan781and Armenia, loitering munitions were used. They were able to,782as you know, basically unmanned aerial vehicles, they moved783into a particular kill box, identified targets that were there.784Then without a human in the loop, they were able to identify785the types of systems that were there, whether it was a tank and786an armored personnel carrier, a command center, a radar station787aircraft, and so forth.788 But because they had that capability, they could then789choose which weapon system based upon which drone was there in790the area and at an appropriate time attack each of them. Is791that the type of--can you talk about, is that what you mean792when you say right-sizing in terms of having the capability for793that particular mission set? Or share with me what you mean by794that.795 Mr. Ferris. Yes, thank you, Senator. In that context, I796think when we talk about right-sizing the model, we're talking797about making sure we're bringing the appropriate solution to798the use case. So, to use your example, we would be looking at,799you know, how the models are used to analyze all that multi-800source information that's coming into the system and from801various sources, but also potentially from different sensors802and systems.803 I think what's important is that we would suggest that by804analyzing, using artificial intelligence to analyze all of that805data, it allows you to elevate the level at which a human can806make that decision. We would still suggest that the human AI807interface is important, and that should be maintained during808these types of operations. But really what AI allows you to do809is to elevate that decision and make it closer to when it needs810to be taken, potentially.811 Senator Rounds. I'm going to--you're following right into812what my next question was going to be, and that is with regard813to--and I'm going to run this all the way down the line again,814but I want to talk a little bit about humans on the loop, and815humans over the loop, and defining each of them, if you would,816in terms of where we're at today and where we're going to be817tomorrow.818 I'm going to talk about it in both offensive and defensive819capabilities. The example that I would use that if you could820buildupon, is we have systems right now that for defensive821capabilities, we arm them, but once they've been armed, they822can automate to protect our platforms.823 That means if you have incoming missiles, particularly if824you're talking, you know, less than a minute to respond, to be825able to identify a missile incoming, such as what we've seen in826the Red Sea region with regard to Houthis attacking our827systems.828 But to be able to identify it, identify the type of weapon829system necessary to take it out, and then to be able to execute830and then to have backups along with it, how far along are we,831and what will AI do with regard to having that whether there's832a human directly in the loop of making that decision, or on the833loop having armed it, or over the top of the loop, not engage834at all.835 I'd like your thoughts, then I'm going to ask our other two836members here as well for their thoughts.837 Mr. Ferris. Yes. Thank you, Senator. So, obviously, I would838say that, currently, we're supporting or we're seeing AI839deployed in an environment with humans in the loop, as you840described, and on the loop where there's some oversight. But841certainly, I don't think we're yet at that over the loop where842they're elevated outside of the analysis and execution of the843mission set, if you will. But, certainly, as agentic AI becomes844more advanced, and the models improve, and become more precise,845and relevant, which is happening at an incredible pace, I would846say we'd be able to see some of that.847 But again, our position at Cohere would be that we want to848work--we would develop--because we deploy models, you know,849with our customers in their environments, we would suggest that850that integration on the front end with the customer and with851our partners having that partnership in development,852deployment, and then, you know, ultimately the decisions in how853those guardrails are put in place. I think that's important on854the front end of really understanding where in that loop it's855necessary to have the human placed.856 Senator Rounds. Mr. Tadross?857 Mr. Tadross. The way that I would kind of look at this is858for human in the loop. What you're sacrificing is speed over859the oversight required to ensure that you're rendering it. In860those cases, I think in, on, or over the loop, it really comes861down to the use case and the speed at which you have to make862the decision.863 So, if the use case is such in a defensive manner, similar864to like a CIWS [close-in weapon system] or an Aegis Cruiser,865which if certain triggers are hit, you default to the machine's866knowledge because the speed at which things are changing is so867great that you can no longer support the decisionmaking868process.869 I think what it comes down to with that's a heuristic-based870system where it's like very clear triggers to be able to871implement that same type of approach with AI would require a872certain amount of evaluation of those systems.873 So, going back to the benchmarking question from earlier,874it would also require having a data infrastructure layer in875place to be able to retrain those models effectively when the876environment changes significantly. As a result of doing that,877you can ensure that this rapid iteration of retraining, and878testing, and evaluation can occur that would still provide the879commander the opportunity to make that informed decision about880if the staff needs to be in on or over the loop.881 Senator Rounds. Thank you. Mr. Mitre? I apologize, am I882saying your name correctly? Is it Mitter?883 Mr. Mitre. Mitre.884 Senator Rounds. Mitre.885 Mr. Mitre. Mitter is fine, too, though. We get it all the886time. Not a problem.887 Senator Rounds. Thank you.888 Mr. Mitre. Yes, no worries, Senator. On this point, I think889fundamentally what the Department of Defense is looking for are890weapons systems and military systems more broadly that are891effective. So, the question is, what is effective in a892particular use case in particular context?893 Now, certainly as the technology progresses, there are more894opportunities to use it in different ways, and along with that895can come greater dependence on the technology. With greater896dependence, you potentially open up new vulnerabilities and new897risks associated with that. So, it's incredibly important to898understand what are ways in which it could go sideways.899 What are some of the vulnerabilities there? When you're900integrating in a broader weapon system where it might act in901ways that are inconsistent with human intentions, and do you902have the right safeguards put in place to guard against those903cases? Are there kill switches that might be necessary? Are904there ways in which you're dealing with a model that's breaking905out of the box and engaging more with the cyber world? Are you906able to cut it off from certain applications if you need to?907 I think it's helpful for the Department to think through908the wide range of potential applications here, and then make909sure that it's thought through how you ensure effectiveness910despite different ways in which the model could react in a911particular context.912 Senator Rounds. Thank you. Senator Rosen.913 Senator Rosen. I want to talk about energy limitations, but914I'm not going to ask this as a question. I'm just going to make915this as a general statement, philosophically. Because if we916move to no humans in the loop, why not just create a grand917video game and save lives? Because at the end of the day, if918it's the AI making the choice, there's still people on the919ground. All of us. Not just men and women in the military, but920the rest of us that live in the world that the computer may or921may not really care too much about.922 So, it's a bigger philosophical question as we move923forward. Not expecting it to be answered here, but in a way, we924have to be sure that we think about that because for every925action these computers might take to each other, theirs versus926ours, the fallout happens to us living here on earth. That's927all I'm going to say. But we got to speak about living here on928earth.929 We got AI energy limitations. You know, a lot of data930centers in Nevada. Let me tell you, there's an increasing931demand for energy. They just gobble it up, and it's a hardware932problem, software problem. It's largely based of course, on the933current architectures that we have.934 Like I said, Nevada's dry weather and our vast open spaces935that we have really become a national leader in data storage936centers. Our companies are constantly innovating, but we know937that the growing use of all this is going to create great938energy burdens on our commercial, our Government Data Centers.939 So, I guess we'll go this way. We'll start with Mr. Ferris.940How do we address this challenge? Do you see it as a barrier to941more widespread DOD and Government adoption? What research,942what should we be investing in to try to maybe reduce that that943great energy suck as it's going to take everything it can,944right?945 Mr. Ferris. Yes. Thank you, Senator. So, Cohere, this is946actually fundamental to our company. We build custom models947designed to be efficient and deployable in the environment that948our clients and customers are working in. So, in pursuit of949that efficiency, a couple of things. One, we're chip agnostic950and cloud agnostic. So, that means we've had to focus on951building our models in somewhat of a resource-constrained952environment. So, we've built----953 Senator Rosen. What if you put it on tanks? You've got954heat, you have to be sure that they adapt in heat environments955and they're going to generate energy, right?956 Mr. Ferris. Absolutely, Senators. But we've built some of957these models to be deployed on as small as two GPUs [graphics958processing unit] or even, you know, we're pushing toward edge959deployments in laptops. So, being able to bring down that960energy cost, but also the infrastructure as a whole. Then, even961it has implications, broadly speaking, into the supply chain as962well.963 Senator Rosen. Thermodynamics. Thank you. What can we do964about all the energy we need to do all of this and then make it965portable?966 Mr. Tadross. Yes, ma'am. So, the way that I kind of look at967this is as these technologies start to be fielded, there's968always an interest in the Department of Defense in order to be969able to operate in a disconnected environment.970 So, what that requirement's going to come along with is971fine tune smaller models that can interact together, which is972similar to the approach that we're taking with INDOPACOM and973EUCOM for agentic warfare. So, what this really results in is a974lower power requirement because back at home station, while975we've been doing the development and training, we're able to976tune these models. You've been using very specific data sets.977So, individual models are very good at a specific thing.978They've been tested and evaluated, and then the interaction979between those models is what can be fielded at the edge. So,980that minimizes the energy requirements as these things begin to981get fielded and proliferated.982 Senator Rosen. Thank you. Mr. Mitre?983 Mr. Mitre. The only thing I'll add is that it's important984to think about the entire tech stack to include power. Not just985the data layer and compute layer, and then, the models itself986and certain applications.987 So, you're right to think holistically. The power is a big988part of that, and certainly, there are ways to find smaller,989more efficient models that you could deploy abroad along the990lines of what the other panelists said. It's worth the991Department looking at that aggressively.992 Senator Rosen. Thank you.993 Senator Rounds. Same question for all of you now. You all994work with the Department of Defense probably in different ways,995but my question is, what can the Department of Defense do with996regard to either policy acquisition policies the way that they997treat contractors? What can they do to enhance their ability to998take advantage of the private sector's capabilities that999they're not doing today? Mr. Ferris.1000 Mr. Ferris. Thank you, Senator. The first thing we'd say is1001we believe that the Department needs to have an adoption1002mindset. We've seen a really good shift. You know, the software1003acquisition pathway and the use of other transaction1004authorities from an acquisition perspective. There are some1005really great strides in acquisition.1006 I would offer using existing mechanisms. I'm an advocate1007for the simple acquisition threshold being, you know, either a1008provision similar to what we have currently. The simple1009acquisition threshold is $250,000 for, you know, contracting1010officer can buy anything under that without a competitive1011process.1012 There's a provision for contingency operations or cyber1013defense and CBRN [chemical, biological, radiological, and1014nuclear] defense, where that simple acquisition threshold is1015raised because of urgent operational requirements. I think1016similarly, we could have an approach in procurement where for1017artificial intelligence, urgent operational requirements,1018perhaps the simple acquisition threshold could be a provision1019for that.1020 What that would do is it would shift the burden away from,1021you know, the DIUs, and DARPAs, and organizations like that1022that are well versed in using OTAs [other transaction1023agreements] and allow contracting officers and project managers1024at like much lower levels in the department to execute and1025acquire these types of capabilities.1026 Senator Rounds. Mr. Tadross?1027 Mr. Tadross. Thank you, Senator. So, when I think about1028making it easier to acquire this technology, I tend to actually1029go back to the AI infrastructure standpoint. The reason for1030that is it actually opens the barrier, reduces the barrier of1031entry of companies to come in. If they're able to operate off1032of a central data repository, then that that company's pathway1033to being able to create relevant technology for the Department1034of Defense is considerably easier than one of the legacies that1035have been in that space for a while and may have troves of data1036that they've saved over 20 years of conflict.1037 Senator Rounds. Thank you. Mr. Mitre?1038 Mr. Mitre. I agree with the panelists on everything that1039relates to narrow AI or AI that exists today. What I think is1040principally lacking from the Department's approach to the issue1041is anticipating where AI might be in a couple of years' time,1042and really working closely with the technologists that are at1043the forefront of developing generative AI and frontier AI1044models to get their head around what that world might look1045like.1046 So, there's a lot of attention, rightfully put toward1047maintaining our lead in the development of technology itself to1048better promote its development, to better protect our lead1049through expert controls, and AI security, and things of that1050nature. But how well does the Department really understand what1051capabilities it may unearth in the next 2, 3, 4, 5 years, I1052don't know, and what that means for the future character of1053warfare. That's crucially important, especially as the1054Department now embarks on developing a new defense strategy.1055 Senator Rounds. One last question for all of you, and you1056don't have to spend a lot of time on this. But is there a place1057somewhere, a safe space, so to speak, where industry and DOD1058can actually interface and ask questions of one another, offer1059ideas, offer products, and so forth that is ongoing? Or is it a1060case-by-case basis?1061 In other words, if industry has a particular product that1062they think would be great in its application within DOD, do1063they know where to go to get it? DOD on the other hand, do they1064have a place where they can go and ask the questions about what1065do you have that can help us fix this problem? Does that exist1066today? Don't everybody speak at once?1067 [Laughter.]1068 Mr. Mitre. Not in a structured and systematic way, right? I1069think it happens in ad hoc cases here and there, but not in a1070coherent approach to really have a tight public-private1071partnership, if you will, to really understand where are we in1072the development of AI technologies relative to key competitors,1073like the Chinese, in particular, what are things that we need1074to be doing to make sure that America maintains that lead.1075DeepSeek is a great example here where surprises like that can1076come out and people wonder, well, what does that mean in terms1077of where we are?1078 I don't think we have that kind of environment to enable1079that constant flow of communication, especially when a cleared1080environment where you can have more sensitive conversations1081with key experts in terms of what's happening with this1082technology and what the U.S. Government needs to be doing in1083partnership with the private sector to maintain America's lead.1084 Senator Rounds. Thank you. Any other thoughts?1085 Mr. Tadross. Yes, Senator. So, I think the closest that1086I've seen of that existing is Project Maven where the efforts1087behind that was to bring technology into the Department of1088Defense in a very aggressive manner. Because they took that1089approach and because you had a single program that was well-1090funded, well organized, and manned by the right individuals,1091what you end up with was a situation in which they were seeking1092to find as many technology experts as they could bring them and1093figure out ways to get them into the Department to satisfy a1094mission requirement that was set forth.1095 Senator Rounds. Thank you. Mr. Ferris, anything?1096 Mr. Ferris. I'll just add that, you know, echo that it is1097very ad hoc and unstructured. However, I think that's precisely1098why actually, you know, people like us end up staying in these1099types of companies and working in them for as long as we do1100because it's important to know those pathways, know those1101venues in which these conversations do unfold, and how to get1102after, you know, getting in front of the Government customer as1103quickly and rapidly as possible, especially when you do think1104you have something that can support the mission. So, it's a1105little bit at this point, it's experience for some of us where1106we can find that opening and get in front of the Department.1107 Senator Rounds. Thank you. Senator Rosen.1108 Senator Rosen. I have one last question. I think for those1109of you who don't know, Maven means ``know it all'' in Yiddish,1110I should say. We should have the Maven marketplace. How about1111that? There you go. That maybe that solves what you need.1112 What I want to talk about and just finish up with, we can't1113do any of this without building our AI workforce. That is1114something that Congress can help invest and promote, and we can1115only go as far as we are willing to invest in all of that. It's1116just so very important.1117 So, for all of you, as we just finish up in our last few1118minutes, the workforce issues that you see in adoption of AI,1119what do we need to do to grow? Well, coders, engineers? All of1120the things that we have to do to build out this robust1121workforce? Because these are the kinds of things that Congress1122does work on and does fund. What advice would you give to us?1123 No one starts in the center. We started on the ends. We'll1124start with you, and I think it's a good way that's something1125that is in our wheelhouse and work on that Maven marketplace.1126Will you? There you go. I'm going to trademark that name. You1127heard it here first.1128 Mr. Tadross. Absolutely, Senator. So, I can say that I'm1129actually very, very proud of the work that we're doing in St.1130Louis. So, in this case, what we're doing is we're taking1131individuals that would normally not participate in the national1132defense and give them an opportunity to support data1133development and AI development in the St. Louis community.1134 So, in some cases, what we've done is taken individuals off1135the fry line, train them on how to look at electro optical1136imagery, gotten them to the point, through training, that they1137are then able to look at synthetic aperture radar, get them to1138the point where they have a clearance, and then even elevate1139them even further so that they're able to pass certain imagery1140tests.1141 Senator Rosen. So, like community college certificate1142programs to bring people just into the workforce, or would you1143say even things like that, right?1144 Mr. Tadross. Yes, ma'am, and give them an opportunity to1145kind of participate in that national defense. This is an area1146where like Scale believes very strongly in. Kind of elevating1147this workforce in order to support the needs of the national1148defense in this space.1149 Senator Rosen. Yes. Perfect. Mr. Ferris?1150 Mr. Ferris. Thank you, Senator. I agree. I mean, I think1151what we would say, we try to partner with, you know, it's a1152public private partnership. That's extremely important.1153Workforce development is critical as part of the body of work1154that the Department and really the Government needs to1155undertake to achieve the advancement in AI that we're hoping1156for.1157 But at within the company, we do partner with educational1158institutions and within the community, and we're searching for1159ways to continue to grow that workforce. I do think it's a1160collaborative process that we need to take with the Government1161and work in concert on it because, from a Cohere perspective,1162we want to be, in terms of our deployment and how we work with1163our customers, it's really early on. So, we want to make sure1164that we're contributing to the workforce development in a way1165that's meaningful for the Department as time goes on.1166 Senator Rosen. Mr. Mitre?1167 Mr. Mitre. This is not exactly my area of expertise, but in1168my experience, there's no more compelling reason to go work in1169Government than for the mission. So, emphasizing that is the1170key ability to attract top technical talent, I think is1171crucial, as is giving them opportunities to develop their1172skills.1173 That requires actually having the right compute1174infrastructure and networking analytic tools available so that1175they can grow and develop their skillset while in Government.1176That's often a challenge to bring together, but there's a1177broader point than just the technical talent, the AI talent1178skillset here as well.1179 Given advances in AI, it's going to impact all elements of1180the workforce. What we're seeing in the private sector right1181now, by way of analogy, is those companies that are better1182leveraging AI or outcompeting companies that don't have it.1183 I think that's likely what we could see in the military1184context, do those militaries that are fully embracing and1185applying it across a range of applications are going to be at a1186significant advantage relative to those militaries that aren't.1187So, I would think a little bit more holistically on the1188workforce dynamics here.1189 Senator Rosen. Thank you. Appreciate it.1190 Senator Rounds. Well, with that, let me take the1191opportunity to thank all three of our presenters here today;1192Mr. Jim Mitre, Vice-President and Director, RAND Global and1193Emerging Risks. Mr. David Ferris, Global Head of Public Sector,1194Cohere, and Mr. Dan Tadross, Head of Public Sector, Scale AI.1195We thank you for participating in this open discussion today1196that's been very, very helpful.1197 My thanks also to my Vice-Chair, Senator Rosen, for1198participating today as well. We appreciate that, and unless you1199have any closing comments, I thank you for being here. Thank1200you for your work, and look forward to continuing to work with1201you and the ideas you have.1202 With that, this Subcommittee hearing of the Cybersecurity1203Subcommittee is now closed.1204 [Whereupon, at 4:29 p.m., the Subcommittee adjourned.]12051206 [all]