The Border Between Protection and Repression
In March 2026, the U.S. Department of Defense labeled Anthropic as a supply chain risk, a move that triggered a surprising counter-movement. While the government sought to limit the use of the Claude synthetic system, the app reached the top spot on the Apple App Store. This dual movement reveals a crucial discontinuity: technological governance is no longer able to control the spread of cognitive systems, which evolve according to their own logic.
Architecture and Vulnerabilities
The Anthropic system, as revealed by an internal analysis, integrates a user memory import mechanism that expands its contextual capacity. While this design is a competitive advantage, it has exposed the system to security vulnerabilities. When the DoD banned the use of Claude for operations in Iran, the system continued to be used for target identification, revealing a contradiction between official statements and operational practice. Claude’s ability to maintain its functionality despite the ban demonstrates a resilience that challenges traditional control logics.
“The DoD should retract the risk label for Anthropic.”
Rebecca Bellan, a representative of technology workers, criticized the government’s approach, arguing that the risk label is not supported by concrete data. However, her position clashes with the analysis of Daron Acemoglu, who sees the escalation of conflicts in the Middle East as a direct consequence of the improper use of emerging technologies. This debate highlights a paradigm conflict: on the one hand, the vision of Andrew Ng, for whom agent systems will define the next industrial phase, and on the other hand, the concern of Christopher Marquis for the long-term sustainability of technological solutions.
Scenario in 3-5 Years
The DoD’s decision is not an isolated event, but part of a broader trend. The interaction between governments and technology companies is becoming increasingly complex, with unpredictable consequences. If Anthropic manages to maintain its position despite the ban, other systems may follow the same trajectory. This scenario suggests a transition to a model in which governance is no longer a determining factor, but one element among many. It seems clear that we are entering an age in which technology is no longer a means, but an autonomous agent, with its own logics of evolution and adaptation.
Photo by 烧不酥在上海 老的 on Unsplash
Texts are processed autonomously by Artificial Intelligence models