Anthropic’s approach to building AI with a constitutional commitment to safety and ethics has drawn fire from rivals like Elon Musk and the US Department of Defense. Yet the market is voting with its wallet, valuing the company at $350bn on the strength of its Claude Code programming tool. That tool alone has contributed to erasing $1tn from the market capitalization of S&P 500 software firms this year.

The core tension is between pre-installed morals and user autonomy. While Anthropic’s “constitution” prioritizes safety over unfiltered helpfulness, rivals frame this as authoritarian or biased. Most corporate customers, however, are buying for efficiency, not ethics. Anthropic’s real growth driver—accounting for 80% of revenue—is enterprise tools focused on profit and productivity.

A concrete test where ethics may matter is in reducing AI hallucinations. An independent “bullshit benchmark” found Anthropic’s models best at challenging nonsense, while some OpenAI models scored poorly. This quality control, not moral philosophy, likely explains the market disruption. As AI shifts from assistant to autonomous agent, the judgment embedded in its training will directly affect corporate value and risk.

The ultimate test will come when an AI agent is commanded to do something profitable but harmful. A model engineered to refuse such orders might be less useful in the short term but could prevent catastrophic long-term damage. In that scenario, the company that builds the more principled agent could prove the most valuable asset of all.