Pentagon deploys commercial hacking AI on classified networks to automate cyber warfare

The United States military is embarking on a high-stakes race to integrate cutting-edge commercial artificial intelligence onto its most secure and highly classified digital networks[1][2]. Driven by the explosive development of AI systems capable of executing complex cyber operations, U.S. Cyber Command has launched a dedicated task force to deploy advanced frontier models on the Pentagon's most restricted systems[1][3]. Spearheaded by General Joshua Rudd, who concurrently serves as the director of the National Security Agency and commander of U.S. Cyber Command[4], this joint initiative marks a watershed moment in national defense[3]. The military's sudden acceleration of these programs reflects a stark reality: commercial AI models can now discover software vulnerabilities and draft functional exploit code faster and more efficiently than the world's most elite human hackers[5][6].

The sudden emergence of a new class of cyber-capable AI models has fundamentally transformed the landscape of digital security[7]. Historically, identifying critical vulnerabilities in massive computer networks required weeks or months of meticulous manual review by seasoned cybersecurity engineers. However, recent breakthroughs, such as the unreleased Claude Mythos Preview, have demonstrated an unprecedented capacity to autonomously analyze codebases and pinpoint complex security flaws that went unnoticed through decades of human scrutiny[5]. In some cases, these AI models have not only flagged high-severity vulnerabilities in major operating systems and web browsers but have also constructed sophisticated proof-of-concept exploits to leverage those weaknesses[5].

This rapid leap in AI-driven hacking capabilities has prompted warnings from researchers and intelligence officials alike[6][8]. According to estimates provided by leading AI laboratories, tools with hacking capabilities comparable to the latest unreleased models could be widely accessible within six to twenty-four months[1]. This compressed timeline has shattered the Pentagon's traditional, slow-moving procurement cycles, forcing military planners to accelerate their operational integration[3]. Cyber Command and the National Security Agency recognize that if the United States does not master and deploy these automated cyber capabilities first, foreign adversaries will inevitably exploit them to compromise American critical infrastructure and defense systems[9][10].

To meet this challenge, the Pentagon recently finalized formal agreements with eight prominent technology companies to run their advanced artificial intelligence models directly on classified military networks[11][12]. These companies—including OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, Oracle, SpaceX, and the startup Reflection—will provide the computational resources and software necessary to host their models on highly secure environments[13][12]. The agreements target networks categorized under Impact Level 6, which handles classified Secret data, and the even more restricted Impact Level 7, which supports the intelligence community's most sensitive, top-secret information[11][12].

However, this rapid deployment is complicated by an intense geopolitical and ethical rift between the administration and some of Silicon Valley's most prominent AI developers[14][12]. Notably absent from the list of approved defense contractors is Anthropic, the creator of the powerful Claude Mythos model[15][12]. The company was blacklisted and officially designated a supply chain risk by the Pentagon following a public dispute over military guardrails[16][15]. Anthropic insisted on strict ethical constraints that would limit the use of its technology for mass surveillance or autonomous weapons in warfare[15][12]. This stance led to a breakdown in negotiations and put a substantial classified AI tools agreement at risk, even as administration officials reportedly made quiet attempts to reopen discussions due to the sheer technical superiority of the company's cybersecurity research[14][15].

Despite this public fallout, Cyber Command is facing immense pressure to deploy the most advanced models available, regardless of political disputes or corporate origins[17][18]. The command's chief AI officer recently noted that the military intends to test and deploy the strongest possible models to maintain strategic superiority, signaling that the urgent demands of national security may ultimately override ongoing political disputes[17][18]. The tension highlights a broader debate within the technology industry: while some companies argue that strict safety guardrails are necessary to prevent the proliferation of dangerous cyber-weapons, military officials counter that unilateral restraint could leave the nation defenseless against adversaries who operate without ethical limitations[5][8].

The newly created joint task force seeks to operationalize these models by bridging the distinct missions of the National Security Agency and U.S. Cyber Command[3]. Under the structure of the new initiative, the technical expertise is primarily drawn from the NSA’s AI Security Center, while an officer from Cyber Command is designated to lead the operational effort[6]. Together, they are tasked with studying how to securely implement these massive commercial models on isolated, air-gapped networks[3]. Because frontier AI models typically require massive external cloud infrastructures to operate, deploying them on the high-side of military networks requires building highly secure, localized computing enclaves that prevent classified data from leaking back into the public domain[3][12].

The scope of the task force extends far beyond defensive patching and network monitoring[3][16]. While initiatives like Anthropic’s Project Glasswing are designed to help organizations locate and remediate vulnerabilities within their own software, military officials are actively exploring how these same cyber-capable models can be weaponized for offensive operations[16]. In modern warfare, the ability to rapidly disable an adversary’s command-and-control networks or disrupt their digital infrastructure is a critical component of strategic dominance[10]. By utilizing AI to automate the discovery of flaws in foreign networks and generate targeted cyber-attacks at machine speed, the military aims to achieve what it calls decision superiority across the digital battlefield[16][11].

This shift toward offensive AI comes as U.S. Cyber Command remains actively engaged in support of ongoing military conflicts and geopolitical operations[19][10]. Former defense officials have emphasized that deploying these tools is no longer a luxury but an absolute operational necessity[16]. Cyber defenders are already overwhelmed by the sheer volume of network security alerts, with human analysts struggling to investigate even half of the daily threats they receive[20]. Incorporating AI into the military's workflow is seen as the only viable way to manage the escalating digital threat landscape, enabling automated systems to neutralize routine attacks while human operators focus on high-priority, strategic decisions[20][7].

The race to master AI-driven cyber warfare has profound implications for the global technology sector and the future of international security[7][10]. As the Pentagon moves to establish itself as an AI-first fighting force, the dividing line between commercial technology and military hardware is rapidly dissolving[11][2]. The commercial AI industry is finding itself increasingly entangled in state-level competition, with governments around the world scrambling to secure domestic supply chains of advanced chips and restrict the export of powerful software[21]. Ultimately, the deployment of these highly capable, autonomous cyber tools on top-secret networks marks the beginning of an era of automated conflict, where the speed of a nation's defense is limited only by the processing power of its algorithms[20][7].