Anthropic was founded by people who did not trust OpenAI's leadership to steward powerful artificial intelligence responsibly, and it now finds itself accused of having intensified the very race it warned about.
The company overtook OpenAI on sales earlier this year and dominates the market for AI coding tools. It filed confidentially for an initial public offering in June 2026, targeting an October listing at a valuation of $965 billion, with Goldman Sachs, Morgan Stanley and JPMorgan Chase named as lead underwriters.
That commercial success is the crux of the critique against it.
Anthropic's stated logic is that reaching the technological frontier is the only way to build the safety protocols capable of preventing catastrophes involving bioweapons or cyberattacks.
The counter-argument is that a well-funded, fast-moving safety lab makes the competitive dynamic worse, not better, by giving rivals something to chase.
Chief executive Dario Amodei is described by colleagues as having believed in the dangers since his early days at OpenAI, when he worked from an air-gapped computer at home and voiced fears about state actors, including a concern about being kidnapped if he travelled to China.
Those warnings predate the technology's credibility, having been raised when GPT-1 and GPT-2 struck most observers as toys.
Whether the same statements now function as marketing is the subject of open debate in Silicon Valley.
Amodei inspires unusual loyalty inside the company, attributed less to conventional leadership technique than to a willingness to make uncomfortable trade-offs and defend them.
He appears to have no wide circle of outside mentors, relying instead on his co-founders, notably his sister and company president Daniela Amodei, and on his wife Camila Clark, a close counsellor who accompanies him to industry gatherings and travelled with him to Miami to meet Jared Kushner and Ivanka Trump.
Clark maintains a very low public profile despite that role.
The company itself has been characterised as bunker-like, with tight security, few leaks and a strong internal culture built around conviction in Claude, its AI model.
Many staff believe the model will drive advances in hard engineering and robotics that reshape global employment within five to ten years, a belief that drives the work rate more than pay does.
The risk in that is obvious enough to have a name inside the industry: a company high on its own supply.
Anthropic has already made compromises, accepting Middle Eastern investment after suggesting it would not, and revising parts of its responsible scaling policy earlier this year.
The pressure comes from the concept of takeoff, the point at which AI can build better versions of itself, and from a widespread belief that whoever gets there first cannot be caught.
That belief, more than any commercial ambition, is what makes the trade-off between safety and speed permanent rather than temporary.
