The clash between national security and the limits of private development
The relationship between the US federal government and leading Silicon Valley firms has entered a scenario of unprecedented tension as a result of the ethical and institutional conflict between the US government and Anthropic. The dispute, caused by the technology company's refusal to make the terms of use of its systems for war and internal surveillance purposes more flexible, exposes the complexity of delegating critical defense functions to deep learning algorithms. This regulatory and corporate clash questions the viability of public contracts for the provision of artificial intelligence software when national security clauses directly collide with the ethical guidelines of the creative companies.
The Pentagon's operational demands: autonomy and biometrics
The core of the technical disagreement lies in two specific requests made by the United States Department of Defense to Anthropic in the context of its military and security operations. First, the US administration requested an exemption from software restrictions to provide full autonomy to unmanned weapons systems, such as tactical drones guided by artificial intelligence, allowing them to execute attacks against predetermined military targets without requiring human approval in each operational phase.
Secondly, the enabling of tools for domestic cyber surveillance and biometric recognition tasks was required. Through this system, federal security forces intended to process video records to automatically identify citizens within the United States, linking faces with civil affiliation data in real time.
The ethical objection and the human control debate
Faced with state demands, Anthropic based its refusal on its corporate principles of security and technological responsibility. The technical dispute is summarized in the theoretical and practical distinction between two models of supervision of automated systems:
Active supervision systems (human-in-the-loop): This approach requires the artificial intelligence to stop its process and request explicit validation and authorization from a human operator before executing an irreversible action, such as launching a projectile or stopping based on facial identification.
Passive supervision systems (human-on-the-loop): In this modality, the system operates with executive autonomy and the human operator fulfills an external supervision role, with the power to intervene only to abort or correct the action after the algorithm has started the decision-making process.
While the Department of Defense requested a transition towards a human-on-the-loop model to streamline tactical response in combat scenarios, Anthropic confirmed that the operational and moral risk of dispensing with direct human intervention in lethal force decisions or civilian surveillance is unacceptable. The firm argued that the lack of explainability and the intrinsic margin of error of recognition algorithms—which could lead to erroneous attacks or biometric false positives—make binding human participation essential at all times.
The use of state power and the unilateral response of the Executive
Faced with the refusal to relax, the United States government escalated the conflict by applying federal public procurement regulations. Covered by supply chain security legislation, the US Executive declared Anthropic as a supplier that represents a potential risk to national security due to operational limitations imposed on its software licenses.
This statement not only deteriorates the company's reputation, but unilaterally enables all federal government agencies to prematurely terminate the million-dollar contracts in force with the technology firm. Given the unilateral breach of these contractual agreements, Anthropic is considering initiating legal action against the State on charges of breach of contract, moving the debate on military ethics to civilian courts.
OpenAI opportunism and the dilemma of shared responsibility
The regulatory fracture between Anthropic and the Pentagon immediately altered the competitive balance in Silicon Valley. OpenAI, the corporation led by Sam Altman and a direct competitor of Anthropic, intervened in the debate expressing its willingness to provide the US government with the requested technological infrastructure, adapting its software to the tactical requirements of the Department of Defense. This corporate positioning highlights the internal fracture of the technological ecosystem: while some firms prioritize the immutability of their ethical codes, others opt for commercial pragmatism and strategic alignment with the State's defense objectives.
Historical background and the maturation of algorithmic ethics
The discussion about the incorporation of automated systems in sectors of high social sensitivity is not recent. Already in 2018, the Buenos Aires prosecutor's office in Argentina positioned itself as one of the four global jurisdictions—along with the United States, China and Estonia—to implement variants of artificial intelligence to optimize processes and responses in judicial files.
The historical experience of these early implementations demonstrates that the value of the technology lies not only in its computational efficiency, but in the design of safeguards that prevent the delegation of human responsibility to the software. The current schism around the use of military artificial intelligence shows that, far from being an abstract technical discussion, data governance and algorithmic autonomy constitute the core of the debate on sovereignty, justice and civil rights in the digital age.