Claude and Grok Are at War in Iran, and Americans Have No Idea

National Interest | Jonathan Guyer

United States military operations in the Middle East are increasingly powered by commercial artificial intelligence models, allowing private tech giants to effectively dictate the rules of algorithmic warfare through their corporate terms of service. This unprecedented delegation of authority occurs as the American public remains largely unaware of the technology's integration into active combat zones.

This shift mirrors the historical expansion of unmanned drone warfare under the Bush and Obama administrations, which similarly lacked legislative oversight and established international guardrails. Today, the Pentagon utilizes the Maven Smart System to run models like Anthropic's Claude and xAI's Grok, despite Maven possessing a low 60 percent accuracy rate, per the Army officer running it. Military targeting errors carry devastating human costs. A missile strike on an Iranian school in Minab killed 175 people, prompting a congressional letter to the Pentagon, while Israel's deployment of the Lavender and Gospel systems in Gaza further demonstrates the high stakes of unregulated, automated warfare.

Comment

The integration of commercial large language models into the Maven Smart System shifts the traditional boundaries of military command and control by outsourcing operational parameters to private terms of service. By relying on platforms like Claude and Grok to process targeting data, military commanders inadvertently subordinate tactical execution to the shifting ethical guidelines of Silicon Valley executives. These corporate policy shifts already manifest in Anthropic's refusal to grant the Pentagon unrestricted use of Claude, forcing reliance on complex workarounds within Palantir's data architecture.

Consequently, the delegation of data-filtering to proprietary algorithms in systems like Lavender and Gospel dilutes command accountability under international humanitarian law. When targeting recommendations are generated by black-box models with a 60 per cent accuracy rate, the legal responsibility for civilian casualties in operations like the Minab strike becomes distributed and untraceable. This accountability vacuum ultimately degrades the command integrity of the US Central Command by replacing human judgment with unverified algorithmic outputs from the Maven Smart System.

Strategic Question for Discussion

Which factor poses a greater threat to command integrity within the Maven Smart System—the 60 per cent error rate of algorithmic targeting or the operational vulnerability of relying on private terms of service from companies like Anthropic?

The available evidence points toward private terms of service posing the more systemic threat, as corporate policy shifts can instantly disable critical operational capabilities without military consent. While a 60 per cent error rate in the Maven Smart System presents severe tactical and humanitarian challenges, these technical deficiencies can be mitigated through human-in-the-loop verification protocols. Conversely, a sudden decision by a private vendor like Anthropic to restrict access to Claude leaves command structures with unresolvable software dependencies during active operations.

Share your assessment in the comments below.

To read in details click here