Yes. Let’s take Anthropic as an example. But my answer would depend heavily on what I was being asked to do inside the company.
One reason I am uncomfortable with Anthropic is the gap between evaluating harm and actually constraining oneself because of it. Christina Wodtke makes this distinction in her critique of Anthropic’s safety policies. Anthropic has built unusually elaborate systems for evaluating catastrophic AI risks, including its Responsible Scaling Policy, capability thresholds, risk reports, and safety roadmaps. At the same time, those policies have been repeatedly revised. Anthropic describes this flexibility as necessary because AI risk is changing quickly. But this raises a question for me: if a company is always able to redefine the threshold as circumstances change, when does an evaluation actually become a constraint?
The recent Colossus deal makes this tension even more concrete. In May 2026, Anthropic agreed to purchase enormous amounts of compute from SpaceX’s Colossus infrastructure, originally built by xAI. SpaceX’s IPO disclosures indicate payments of $1.25 billion per month, potentially continuing through May 2029, although either side can terminate with 90 days’ notice. To me, this further complicates Anthropic’s identity as the “responsible” AI company.
The idea of distance in one of the readings might be relevant here.The text argues that in a deeply interconnected economy, it is nearly impossible to avoid every organization connected to something objectionable. Instead, we can ask how close our own work is to the harm. There is a meaningful ethical difference between directly performing an action I believe is wrong and doing unrelated work for an organization in which that action also occurs. The text also points out that joining an imperfect organization may sometimes give an employee a chance to change it from within.
I find this persuasive partly because of my own experience at UWC. UWC presents itself as a deeply idealistic institution, and I genuinely believed in many of its values. Yet I also saw situations in which the need for funding led the institution to use students in ways that I found ethically questionable. That experience made me skeptical of the idea that organizations can simply choose between being “ethical” and “unethical.” Even organizations founded around admirable ideals have to acquire money, compete for resources, satisfy stakeholders, and survive.
From a sociological perspective, this relates to the problem of alienated labor under capitalism. Large organizations divide production into so many roles that responsibility becomes dispersed. A person may contribute to a system without ever individually intending its larger consequences. At the same time, companies face structural pressures to grow, compete, reduce costs, and secure scarce resources. This does not make harmful practices morally acceptable, but it helps explain why simply replacing “bad people” with “good people” rarely solves the problem.
My question will be: Can a company remain economically viable while building institutions strong enough to constrain its own incentives?
