Pentagon Approves Grok for Classified Systems as Anthropic Faces Ultimatum

Pentagon Approves Grok for Classified Systems as Anthropic Faces Ultimatum

In a landmark development with profound implications for national security and artificial intelligence ethics, the U.S. Department of Defense has officially entered into an agreement with Elon Musk’s xAI to integrate its large language model, Grok, into classified military systems. The move, confirmed by a senior Defense official to Axios, ends Anthropic’s exclusive access to the Pentagon’s most sensitive AI infrastructure and signals a decisive pivot toward models with fewer ethical constraints.

According to Axios, Grok will now be authorized for use in intelligence analysis, weapons development, and battlefield decision-support systems — domains previously reserved exclusively for Anthropic’s Claude model. The shift comes amid a deepening rift between the Pentagon and Anthropic, which has steadfastly refused to remove its safeguards against mass surveillance of U.S. citizens and the development of fully autonomous weapons systems. Defense Secretary Pete Hegseth is set to meet Anthropic CEO Dario Amodei for what sources describe as an ultimatum: comply with the Pentagon’s demand that Claude be made available for “all lawful purposes,” or face designation as a “supply chain risk” — a classification that could trigger contract termination, exclusion from future procurements, and reputational damage across government and private sectors.

The implications of this decision extend far beyond corporate contracts. Grok’s deployment in classified environments raises urgent questions about accountability, bias, and the erosion of ethical guardrails in military AI. Unlike Claude, which has publicly committed to ethical boundaries — including blocking use cases that violate U.S. civil liberties — xAI has not disclosed specific limitations on Grok’s functionality. Axios reports that xAI agreed to the Pentagon’s broad “all lawful use” standard, a phrase that, in military context, leaves wide interpretive latitude. While “lawful” may technically exclude violations of international humanitarian law, critics warn that such language lacks enforceable constraints and could enable uses that skirt ethical norms under the guise of legal permissibility.

The Pentagon’s pivot also underscores the growing commercialization of AI for warfare. Google’s Gemini and OpenAI’s ChatGPT are already accessible in unclassified defense networks, with Google reportedly nearing a deal to bring Gemini into classified systems, according to The New York Times. OpenAI, by contrast, remains hesitant, likely due to internal ethical review boards and public pressure. But with xAI’s entry, the military now has a powerful, politically aligned alternative to Anthropic’s more cautious approach. Sources indicate that Grok was already tested in simulation environments tied to the recent Maduro raid in Venezuela — an operation previously reliant on Claude via Palantir’s AI integration platform.

Anthropic’s resistance has been rooted in its founding principles. Co-founded by former OpenAI researchers, the company has positioned itself as a moral counterweight to unchecked AI proliferation. Its refusal to compromise on surveillance and lethal autonomy reflects a broader industry debate: Can AI be both powerful and principled? The Pentagon’s response suggests a resounding “no” — prioritizing operational flexibility over ethical oversight. If Anthropic declines to capitulate, it risks being sidelined in a $100 billion-plus defense AI market, potentially forcing its clients — including allies and intelligence partners — to seek alternatives.

Meanwhile, Congress is reportedly preparing hearings on AI ethics in defense, with bipartisan concern growing over the normalization of unregulated military AI. Advocacy groups like the Center for AI Safety and the Electronic Frontier Foundation have condemned the Grok deal as a “dangerous precedent.” As the U.S. military accelerates its AI integration, the line between innovation and irresponsibility grows perilously thin.