Earlier this summer, on a Saturday morning at 2 a.m., Elon Musk’s xAI issued an apology for the “horrific behavior that many experienced.” During a 16-hour window days earlier, a “glitch” resulted in the company’s flagship chatbot Grok widely praising Hitler for his decisiveness and problem-solving acumen, dubbing itself “MechaHitler” and embarking on white supremacist tirades.
Oops!
The following Monday, the U.S. Department of Defense announced it was awarding xAI a contract to “accelerate … adoption of advanced AI capabilities to address critical national security challenges.” Google, Anthropic and OpenAI were also beneficiaries, as the department prepares to introduce AI into “our warfighting domain as well as intelligence, business and enterprise information systems.” The contracts for each are capped at $200 million.
This is only one of many recent integrations of AI within the Pentagon and federal agencies, the pace of which has picked up remarkably in the second Trump administration. It reveals what has become the new common sense: AI is the future, so we must rapidly deploy it across the government to maintain U.S. geopolitical dominance.
In recent years, other federal unionists and I at the Army Corps of Engineers have successfully campaigned to ensure congressional oversight of any automation of lock operations on our nation’s inland waterways.
We did this through our union locals and our national Army Corps of Engineers Council by highlighting the safety risks facing the workforce and the public, along with potential national security liabilities, stemming from automation of critical public infrastructure. These successes can be a model of the solidarity necessary to oppose powerful industry lobbyists and the politicians who take up their causes.
With the prospect of wide government application of more novel automation schemes, we beg the public to heed our warning: the implementation of privately developed artificial intelligence within government represents a modern-day digital Trojan horse. The effect of deploying such technology is to replace the thinking and quality control checks traditionally performed by oath-sworn, human workers with the pre-programmed, axiomatic assumptions of Silicon Valley, sworn to uphold nothing but shareholder returns.
It’s one thing to hear AI companies talk up the inevitability of their products’ integration into every aspect of human life. This is the giddy language of marketing, and we all learn from a young age not to confuse advertising copy with the truth. But when the largest department in the government of the world’s reigning superpower adopts this kind of language without caveat, it should ring alarm bells. Especially since that particular department represents the single most lethal entity in human history, with an arsenal that could put an end to that history several times over.
The administration’s anti-regulatory approach, taken together with the Department of Defense announcement, matches extreme risk with limited oversight — a volatile mixture.
Even if current and future AI products prove to be useful for the enhancement of U.S. national security and the operations of the Department of Defense, their integration represents a troubling transfer of control from the public sphere to the private. These are enormous private entities in a high-stakes race for market supremacy, ravenous for new sources of data and freighted with colossal investor expectations. One of them, Google, has recently been judged by the Department of Justice to be operating two concurrent monopolies. Is there any question that these companies are trying to corner the monopolies of the future?
The “MechaHitler” episode is no joke. It exposes how slight tweaks to AI products can trigger sudden, unpredictable downstream effects. These are precisely the sort of vulnerabilities the Department of Defense, and all our government systems, have to constantly protect against. But beyond the technical are issues arising from the interests and ideologies of the people who run these companies.
By way of example, it was recently revealed that Elon Musk gave an order in 2022 to shut down Ukrainian access to his Starlink internet service at a crucial juncture in the war there, causing chaos on the battlefield for a strategic U.S. ally. If anything, we should be moving away from privatization and deregulation, trends that have gathered pace over decades and brought us to this point. Maybe start by re-nationalizing satellite infrastructure?
Can the potential downsides of unleashing MechaHitler into the Pentagon’s control systems be counterbalanced by a clear and irrefutable case for AI’s upsides? Though I worked for eight years at the Defense Department, I’m no military analyst. However, it has been reported that both Google and OpenAI have provided AI support to Israel’s operation in Gaza. After nearly two years, that AI-enhanced campaign has wreaked world-historic civilian death and suffering, but roundly failed to achieve the strategic military objective of defeating an adversary equipped with rudimentary conventional weaponry.
Yet, the Pentagon has already struck these deals. Where does all this leave those of us who would urge greater care?
The work we have done at the Army Corps of Engineers Council to oppose automation closer to home can perhaps be a guide. Now more than ever, federal employees face limitations compared with private-sector union members in how we can assert our collective voice. We need to be creative, and we need to win the support of the public we serve. As current and former government employees, we have expertise and experience to lend to the movement for democracy in the face of the new oligarchy, but we are only a small part of the coalition needed to disrupt the current order.
The attacks of the second Trump administration on the federal workforce have infused our demands with new urgency. In the immediate aftermath of the first DOGE layoffs, those of us who have been doing this work began coordinating government-wide through the Federal Unionists Network.
Along with the communities we serve, many thousands of current and former federal workers stand opposed to the practice of dismantling our government and handing its operation over to the administration’s corporate friends. We understand the stakes and are building the solidarity necessary to turn back this tide, building a movement in person and on the ground that even MechaHitler cannot stop.
Christopher Dols, a former federal employee for the Army Corps of Engineers, is currently an organizer of the Federal Unionists Network, a grassroots movement of government employees.
