The US Department of Defense is revising its classified AI deployment doctrine, planning to grant algorithms broader authority in decision-making and target selection. This strategic shift has already led to a major conflict with Anthropic, whose ethical constraints on Claude models became an obstacle to fulfilling defense objectives.
What Happened
The Pentagon is implementing technologies to automate intelligence processing, including the automatic downgrading of classification levels from Top Secret to Secret, and using AI to manage artillery strikes. As a result of this process, the department labeled Anthropic a "supply chain risk" due to the strict ethical safeguards in Claude models, leading to the termination of a $200 million contract and a subsequent lawsuit from the company.
Context
As part of a strategy to accelerate AI adoption, the defense department aims to minimize bureaucratic barriers as much as possible. There is a visible reorientation of government procurement toward providers such as OpenAI, Google, and SpaceX, who are willing to provide tools with fewer restrictions on combat application.
Why It Matters for the Industry
The Anthropic case sets a precedent where model alignment (ethical tuning) may be viewed by government clients as a technical risk to operational efficiency. This could lead to a fragmentation of the AI market into two parallel branches: civilian (safety-first) and defense (capability-first/unfiltered), where the priorities will be inference speed, accuracy, and the absence of "ideological constraints."
Why It Matters for Users
For AI market participants, this is a signal that the arms race is transitioning from the "smart assistant" phase to the phase of direct combat application tools. This opens niches for the development of specialized "unfiltered" solutions and open-weights models that can be fine-tuned for specific military tasks without external control from the developer.
What Is Not Yet Known / Limitations
Legal experts point to high litigation risks and the need to challenge the "supply chain risk" designation, while developers are focusing on finding new niches for unfiltered solutions.
Sources
Author
Look at AI, Editorial Team