Red Lines in the Rubble: OpenAI Enters the “Department of War” as Claude AI Powers Strikes on Iran

Just as the United States Department of Defense and AI startup Anthropic entirely severed ties over the “red lines of model usage,” OpenAI recently announced the forging of an agreement with the military to deploy its advanced AI systems within classified environments. Astonishingly, OpenAI emphasized that it has not dismantled its models’ safety guardrails; rather, it has explicitly delineated “three red lines of usage.” Concurrently, OpenAI CEO Sam Altman publicly voiced his vehement opposition to the US government’s classification of Anthropic as a “Supply Chain Risk.”

In contrast to the two immutable baselines previously staunchly defended by Anthropic (the prohibition of mass domestic surveillance and fully autonomous weaponry), the stipulations OpenAI presented to the Defense Department are markedly more specific, etching three insurmountable red lines:

• Absolute prohibition on mass domestic surveillance.
• Absolute prohibition on fully autonomous weapons systems.
• Absolute prohibition on high-risk automated decision-making systems (such as profound discretionary frameworks akin to “social credit” mechanisms).

The preceding collapse of negotiations between Anthropic and the military stemmed from the latter’s insistence on a clause permitting “any lawful use,” coupled with a refusal to countenance additional technical restrictions. Anthropic surmised that conceding to such terms would entirely strip the enterprise of its dominion over the model’s practical applications.

How, then, did OpenAI persuade the military to acquiesce? The linchpin lies in the “deployment architecture.” OpenAI adopted a strategy restricted to cloud deployment, preserving a “Safety Stack” that is independently operated and perpetually updated by OpenAI itself, alongside the orchestration of joint oversight by engineers possessing requisite security clearances. Through this dual safeguard of technical and contractual frameworks, OpenAI succeeds in penetrating the national classified apparatus while tangibly retaining its regulatory grasp over the model’s utilization.

OpenAI CEO Sam Altman subsequently unveiled further nuances on the social platform X. He articulated that OpenAI initially intended solely to participate in unclassified initiatives, and had even previously rebuffed the very classified contract that Anthropic had briefly entertained. However, the precipitous deterioration of the landscape this week compelled OpenAI to accelerate its negotiations with the Defense Department. Altman noted that the military exhibited profound “flexibility” regarding the security architecture proposed by OpenAI. He stressed that the primary impetus behind this hastily forged pact was to “de-escalate the situation,” harboring the hope of establishing a paradigm that guarantees all future American AI laboratories can render their services to the military under identical security prerequisites.

More crucially, both OpenAI officially and Sam Altman personally levied scathing critiques against the Defense Department’s designation of Anthropic as a “Supply Chain Risk.” They contend that affixing a label—typically reserved for adversarial nations such as China or Russia—upon a premier, domestic AI enterprise is a profoundly detrimental maneuver, yielding zero benefit for either the American AI industry or national security.

It is noteworthy that amidst this maelstrom of controversy, both Anthropic and OpenAI deliberately invoked the archaic historical moniker “Department of War” (DoW)—a title retired prior to the 1947 reorganization—in their official proclamations, eschewing the contemporary and ubiquitous “Department of Defense” (DoD). External observers interpret this calculated rhetorical choice by the tech behemoths as a potent insinuation, highlighting that the deployment of these AI models has already encroached upon the realm of “substantive military operations and lethal battlefield applications,” transcending mere administrative defense protocols.

And the stark reality aligns precisely with this assessment. According to reporting by The Wall Street Journal, within hours of President Trump announcing a comprehensive ban on Anthropic, the US military continued to leverage Anthropic’s Claude AI technology during airstrikes directed against Iran. Multiple vanguard units, including the United States Central Command, remain profoundly reliant upon Claude for intelligence analysis, target acquisition, and the simulation of battlefield scenarios.