Pentagon AI 'Kill Chain' - Artificial Intelligence in Warfare

AI and the US Military: A Balancing Act

Prominent artificial intelligence firms, including OpenAI and Anthropic, are navigating a complex situation as they seek to provide software solutions to the United States military. The core objective is to enhance the Pentagon’s operational efficiency, while simultaneously ensuring their AI technologies are not utilized for lethal purposes.

Current Applications of AI in Defense

Currently, these AI tools are not deployed as direct weapons systems. However, Dr. Radha Plumb, the Pentagon’s chief digital and AI officer, indicated to TechCrunch that AI is providing the Department of Defense with a “significant advantage” in threat identification, tracking, and evaluation.

Dr. Plumb further explained that AI is accelerating the “kill chain,” enabling commanders to react more swiftly to safeguard forces. This acceleration is achieved through improved processing and analysis of critical data.

Understanding the "Kill Chain"

The “kill chain” represents the military’s systematic approach to neutralizing threats. It encompasses a sophisticated network of sensors, platforms, and weaponry. Generative AI is demonstrating particular value during the initial stages of the kill chain – planning and strategic development – according to Dr. Plumb.

Evolving Policies and Partnerships

The collaboration between the Pentagon and AI developers is a relatively recent development. In 2024, OpenAI, Anthropic, and Meta revised their usage guidelines to permit U.S. intelligence and defense agencies to utilize their AI systems. Despite this change, a firm prohibition remains against the use of their AI to inflict harm on humans.

“We’ve been really clear regarding the permissible and impermissible applications of their technologies,” stated Plumb, describing the Pentagon’s approach to working with AI model providers.

A Surge in Collaboration

This policy shift has spurred a period of intense engagement between AI companies and defense contractors.

• Meta established partnerships with Lockheed Martin and Booz Allen to integrate its Llama AI models into defense agency operations in November.
• Anthropic simultaneously collaborated with Palantir.
• In December, OpenAI forged a similar agreement with Anduril.
• Cohere has also been quietly deploying its models through Palantir.

Potential for Policy Changes

As generative AI continues to prove its utility within the Pentagon, it may incentivize Silicon Valley to further relax its AI usage policies, potentially allowing for a broader range of military applications.

“Generative AI is particularly helpful in simulating various scenarios,” Plumb noted. “It facilitates the optimal utilization of all available tools for commanders, while also encouraging creative thinking regarding response options and potential trade-offs when facing a potential or series of threats.”

Ethical Considerations and Policy Compliance

The specific technologies currently employed by the Pentagon for these tasks remain undisclosed. However, even utilizing generative AI in the early planning phases of the kill chain appears to potentially contravene the usage policies of several leading AI model developers.

For instance, Anthropic’s policy explicitly prohibits the use of its models to create or modify “systems designed to cause harm to or loss of human life.”

In response to inquiries, Anthropic directed TechCrunch to a recent interview with its CEO, Dario Amodei, in the Financial Times, where he defended the company’s involvement in military projects.

OpenAI, Meta, and Cohere did not respond to TechCrunch’s requests for comment regarding this matter.

AI Weapons and the Question of Life and Death Decisions

A significant discussion within the defense technology sector has recently emerged, centering on the ethical implications of allowing AI-powered weapons to independently determine life or death outcomes. Claims have been made suggesting the U.S. military is already deploying such systems.

Palmer Luckey, CEO of Anduril, recently highlighted on X that the U.S. Department of Defense has a well-established record of acquiring and utilizing autonomous weapon technologies, citing systems like the CIWS turret as examples.

Luckey stated, “For decades, the DoD has been procuring and deploying autonomous weapons systems. Their application – and even international sale – is comprehensively understood, governed by strict regulations that are not optional.”

However, when questioned by TechCrunch regarding the Pentagon’s purchase and operation of fully autonomous weapons – those functioning without human intervention – Deputy Secretary of Defense Kathleen Hicks dismissed the notion.

“The concise response is no,” Hicks affirmed. “Both for reasons of dependability and ethical considerations, human involvement will always be maintained in the decision to utilize force, encompassing all our weapon systems.”

The term “autonomy” itself lacks precise definition, fueling industry-wide debates regarding the point at which automated systems – including AI coding tools, autonomous vehicles, and self-activating weaponry – achieve genuine independence.

Hicks characterized the concept of automated systems independently making critical decisions as overly simplistic, asserting that the reality is far removed from “science fiction.” She proposed that the Pentagon’s integration of AI involves a collaborative dynamic between humans and machines.

“There’s a tendency to envision robots operating independently, generating recommendations that humans merely approve,” Hicks explained. “This does not reflect the nature of human-machine collaboration, nor is it an optimal approach to leveraging these AI systems.”

Understanding Human-Machine Teaming

The Pentagon views AI integration not as replacing human judgment, but as augmenting it. Senior leaders remain actively engaged in the decision-making process throughout the entirety of operations involving these technologies.

This collaborative approach ensures that ethical considerations and strategic objectives are consistently prioritized, preventing fully autonomous operation in critical scenarios.

• Key takeaway: The U.S. DoD maintains that humans will always be involved in the decision to use force.
• Ambiguity of Autonomy: The definition of "autonomy" remains a point of contention within the tech industry.
• Human-Machine Collaboration: The Pentagon emphasizes a collaborative model, rather than fully independent AI operation.

Artificial Intelligence Safety Concerns within the Pentagon

Collaborations between the military and technology companies have frequently encountered resistance from personnel within Silicon Valley. In the previous year, numerous employees from both Amazon and Google faced termination and arrest following demonstrations against their respective companies’ military agreements with Israel, arrangements designated under the name “Project Nimbus.”

In contrast, the reaction from the artificial intelligence community has been relatively subdued. Certain AI researchers, including Evan Hubinger of Anthropic, posit that the integration of AI within military structures is unavoidable. They emphasize the necessity of direct engagement with the armed forces to guarantee responsible implementation.

Hubinger articulated in a November post on the LessWrong online forum, “A serious consideration of the catastrophic risks posed by AI necessitates engagement with the U.S. government, a profoundly significant stakeholder. Attempting to preclude the U.S. government from utilizing AI is not a practical approach.” He further stated, “Addressing catastrophic risks alone is insufficient; preventative measures must also be taken to safeguard against potential misuse of models by governmental entities.”

Stay informed with TechCrunch’s AI newsletter! Subscribe here to receive it weekly on Wednesdays.