As a longtime professor and leader of the College of Public Affairs at the University of Baltimore, I spend a good deal of time thinking about power – who exercises it, how it’s employed, and whose ox gets gored when it’s constrained. Recent advances in artificial intelligence (AI) provide a classic case study in the politics of power.
The faculty I lead are hardly trying to wall AI off from their classrooms. Indeed, my university took the opposite approach by launching a Center for AI Learning and Community-Engaged Innovation (CAILI) and then purchasing a multi-platform subscription to a panoply of AI services for instructional and educational purposes.
We are teaching our students – future public servants, policy analysts, and nonprofit leaders – to use AI responsibly and transparently. We require disclosure and demand verification. We insist the final judgment, especially in matters of public affairs, remains human. At the same time, we acknowledge the obvious: AI platforms are transforming the skills that students must acquire, how scholarly inquiry is conducted, what knowledge is conveyed, and how ethics are viewed in a rapidly changing professional landscape. The shift isn’t theoretical. It’s here.
But when it comes to AI, casting a cold eye on the motivations of those speaking loudest – and lobbying hardest – amounts to nothing more than basic prudence.
Few companies have framed the AI debate in moral terms as forcefully as Anthropic. Its chief executive, Dario Amodei, has repeatedly warned of “civilizational concerns” and the possibility that advanced AI systems could pose existential risks. The firm presents itself as a safety-first laboratory – an institution born from dissatisfaction with the industry’s speed and bravado. In a widely read January 2026 essay, Matteo Wong of The Atlantic described Anthropic as the “superego” of the AI industry, torn between its conscience and its competitive instincts.
The phrase lingers because it captures something real. But it also invites a deeper question: what happens when the industry’s conscience is one of its most aggressive innovators? Simply put, has Anthropic become the id of the industry, developing AI models capable of unleashing the ills it has long preached against? Are we simply to trust Anthropic because they paint themselves as the company that cares?
Anthropic’s own constitution acknowledges what the company calls its “peculiar” position. It warns that sufficiently powerful AI may be extraordinarily dangerous, yet argues that its own safety-oriented laboratories should be the only ones entrusted to stretch AI’s frontier. The implication is straightforward: the safest actor - Anthropic - must lead the race.
That sounds reassuring. But is it?
Anthropic is a private company, rooted in the single-minded goal of making its founders into billionaires. A few days ago, Anthropic raised $30 billion in capital from a herd of private equity firms and industrial giants such as Microsoft Corp. Outwardly, Anthropic is structured as a public benefit company, supposedly led by a disinterested board. But are we really to believe that this board, directing an organization valued at $380 billion, will always work in society’s best interests? Can we be sure Anthropic will ignore the will of its major investors if conflicts arise?Forgive my cynicism but I think not. I’m old enough to remember Google LLC’s quaint corporate motto of “Don’t be evil,’’ shortly before it began scraping and hoarding the private data of every soul on earth.
Anthropic is building some of the most capable frontier models in existence. It publishes detailed analyses of how its systems could be misused for insider threats or destabilizing activity, even as it releases increasingly powerful iterations. It openly entertains the possibility – still speculative – that advanced AI systems might one day develop their own moral code, independent of humans. And Anthropic assures us that it alone is uniquely qualified to prevent machines from behaving badly –or snuffing us humans from existence.
But at the same time, Anthropic is building the AI models most likely to bring its doomsday scenarios into reality. Last year, Amodei predicted AI would likely eliminate 50% of the country’s entry-level white-collar jobs in five years, possibly leading to unemployment rates of 20%.
A few weeks ago, Anthropic unveiled an AI model that can run financial analyses, do research, create documents, spreadsheets and all of the digital office machinery currently built by humans. Amodei bragged at a recent forum that his employees now use AI to write 100% of the company’s computer code, a remark that surely sent a shiver up the spines of the estimated 24.3 million coders working on the planet today.
Recently, it emerged that Anthropic is chafing under a $200 million contract with the Pentagon to deploy its AI capability to develop autonomous weapons systems that need no human control. Public reporting suggests Anthropic’s Claude model also was used in a classified mission tied to the capture of former Venezuelan president Nicolás Maduro earlier this year.
Publicly, Anthropic has vowed not to allow its tech to be used for domestic surveillance or robot warfare; the Pentagon responded by saying it would punish the company by designating it a “supply-chain risk.” Such a designation that could lead to business losses for Anthropic far in excess of the Pentagon’s contract.
The irony here is obvious: Anthropic warns of the risks associated with increasingly autonomous or morally salient AI systems. At the same time, it’s the only major frontier lab building AI models possessing this capability. The company warning of AI’s existential risks to mankind is the one beavering away on building these systems.
To be sure, dual-use technology has always blurred moral boundaries. Nuclear physics didn’t not remain in the laboratory; neither did cyber capabilities. Even defensive biomedical research – such as gain-of-function experiments intended to understand viral transmissibility– has demonstrated how safety-justified frontier work can generate global consequences when safeguards falter. Good intentions don’t neutralize systemic risk.
Anthropic has spent millions lobbying Congress to support more AI regulation. The stated rationale is that guardrails are essential to prevent catastrophic misuse. Perhaps but history suggests regulation tends to entrench incumbents and create monopolies.Compliance costs, auditing requirements, and licensing thresholds box out smaller entrants and favor firms like Anthropic with deep capital reserves.
For a company valued in the hundreds of billions of dollars, stringent regulation seems less a brake than a moat. And when the most strident warnings about existential AI risk come from the company best positioned to thrive under tight regulations, it’s reasonable to ask whether altruism and strategic interest are aligned a bit too neatly.
I’m not denying advanced AI carries risk. My concern is the framing of the debate. The company with the massive Pentagon contract is the one shouting loudest about the dangers of militarizing AI. The company warning about catastrophic job losses is the one killing off all the coders and threatening white-collar jobs. Yet we are told to simply accept the assertions from a non-public company that it surely knows best.
Anthropic may well be more earnest in its safety rhetoric than others. But corporate bromides are no substitute for action. When moral urgency aligns seamlessly with commercial profits and geopolitical leverage, prudence demands more than applause. It demands scrutiny.
