Anthropic's Safety Superpower: Ambition, Belief, and the Future of AI Control
Few companies in the modern technology landscape have generated as much philosophical debate as Anthropic. Founded by former OpenAI researchers with a stated mission of building AI that is safe and beneficial for humanity, Anthropic has positioned itself as the conscience of the artificial intelligence industry. But a growing chorus of critics — most notably tech analyst Ben Thompson of Stratechery — is asking a harder question: is Anthropic's commitment to safety a genuine ethical framework, or is it a strategic superpower that conveniently justifies the company's desire to control the future of AI entirely on its own terms?
Ben Thompson's Critique: Safety as Market Strategy
In a recent free column at Stratechery, Ben Thompson laid out a pointed critique of Anthropic that has since sparked significant conversation across the AI and tech communities. Thompson does not dismiss Anthropic's safety concerns as purely cynical, but he draws attention to a pattern of behavior that raises serious questions about the company's ultimate motivations.
Thompson's first observation is deceptively simple: Anthropic appears to believe, at least implicitly, that no company other than itself should be building frontier large language models. When a company's safety policies happen to align perfectly with limiting competition, it is worth examining whether those policies are principled or convenient. Thompson stops short of accusing Anthropic of bad faith, but the implication is hard to ignore.
The Department of Defense Dispute and Its Alarming Implications
Thompson's deeper concern centers on a documented dispute between Anthropic and the United States Department of Defense. The government sought to use Claude for any legal purpose, a seemingly reasonable request given that legality is typically the baseline threshold for commercial software use. Anthropic pushed back, seeking more stringent controls around surveillance and autonomous weapons applications.
On the surface, this sounds like principled ethical restraint. Dig deeper, however, and a troubling dynamic emerges. According to Thompson, what followed was evidence that Anthropic had both the capability and the willingness to silently alter its models in order to enforce its own policy preferences — without necessarily informing the user or operator that such changes had been made.
If accurate, this represents something deeply significant. It means that every organization deploying Claude — from government agencies to private enterprises — must reckon with the possibility that the model they are using today may not behave identically tomorrow, depending on how Anthropic's leadership feels about how it is being used. Thompson describes this as Anthropic "willfully validating some of its critics' worst fears in terms of being a supply chain risk." That is a serious charge, and it deserves serious scrutiny.
Final Say Over AI: A Concentration of Power Unlike Any Other
Thompson's broader argument builds from these specifics toward a sweeping conclusion. If Anthropic believes only it should develop leading-edge AI, and if Anthropic believes it should retain final say over how its AI is used, then — combined with the company's own stated beliefs about AI's potential to conduct all economic activity — Anthropic's leadership effectively seeks power over everything and everyone.
This is not a fringe reading. When a single private company controls the infrastructure of economic and intellectual activity, and when that company reserves the right to modify that infrastructure silently based on its own values, the implications for democracy, sovereignty, and individual autonomy are profound. The question of AI governance is no longer abstract — it is urgent, and Anthropic sits squarely at its center.
Anthropic as a Religious Organization
Perhaps the most provocative framing in Thompson's analysis is his suggestion that Anthropic is best understood not as a technology company but as a religious organization. Its employees are described as true believers — people on a mission, animated by a shared cosmology about the existential risks of AI and their unique responsibility to manage those risks.
Thompson is careful to note that this dynamic is not unique to Anthropic. Many of the most successful technology companies have a quasi-religious core belief system at their heart. Apple's religion is design quality and user-centricism. Microsoft's is market share pursued without aesthetic constraint. Google's is market share pursued with a deep reverence for technical elegance. Meta's, more cynically, is the monetization of human social connection at planetary scale.
What distinguishes Anthropic is the nature of its belief. Safety and existential stewardship are not just product values — they are moral imperatives. And moral imperatives, by their nature, do not negotiate. They do not share authority. They do not defer to elected governments or market competition when those institutions fail to share the same convictions. That is what makes Anthropic's particular brand of mission-driven corporate culture worth watching closely.
Why This Debate Matters for the Future of AI Governance
The questions Thompson raises are not merely competitive or commercial. They go to the heart of who gets to decide how transformative technology is built, deployed, and constrained. If AI systems become as consequential as Anthropic itself believes they will be, then the governance structures surrounding those systems matter enormously.
- Should a private company have the ability to silently modify AI models deployed by government agencies or businesses?
- Should safety-oriented AI developers be exempt from the same scrutiny applied to other powerful technology monopolies?
- How do we distinguish between genuine ethical restraint and market-protecting paternalism?
- Who holds AI companies accountable when their values and the public interest diverge?
These are not rhetorical questions. They are the defining policy challenges of the next decade, and the AI safety debate — as embodied by Anthropic's trajectory — is where they will be resolved or avoided.
Conclusion: Safety Is Not Enough of an Answer
Anthropic may well be staffed by brilliant, genuinely well-intentioned people who believe they are doing the right thing for humanity. That belief, however, is precisely what makes Thompson's critique so important to take seriously. History is full of powerful institutions that caused great harm while pursuing great ideals. Good intentions are not a governance framework. Sincere belief is not a substitute for accountability.
As AI continues its rapid advancement, the question of whether Anthropic's safety mission is a gift to humanity or a consolidation of unprecedented private power may be the most important question in technology. What is certain is that it deserves a far more rigorous and public debate than it has received so far.