One of the AI industry’s most prominent figures has set out his concerns about humanity’s present and future in the emerging technological landscape.
Humanity is entering a phase of artificial intelligence development that will “test who we are as a species”, according to the head of the fast-growing AI start-up Anthropic, who argues that the world must wake up to the risks.
Dario Amodei, co-founder and chief executive of the company behind the popular Claude chatbot, expressed his fears in a 19,000-word essay entitled “The Adolescence of Technology”.
Describing the arrival of exceptionally powerful AI systems as potentially imminent, he wrote: “I believe we are entering a rite of passage, turbulent but inevitable, which will test who we are as a species.” Amodei added that humanity was about to acquire almost unimaginable power, while it remained deeply uncertain whether its social, political and technological systems were mature enough to handle it.
The technology entrepreneur, whose company is reportedly valued at $350 billion, said the essay was an attempt to shake the world out of its complacency and underline the need for action on AI safety.
Amodei published the essay as the British government announced that Anthropic would help build chatbots offering careers advice and job-search support, as part of the wider development of an AI assistant for public services. The previous week, the company released an 80-page document it described as a “constitution” for Claude, explaining how it intended to make its AI broadly safe and ethical.
Amodei co-founded Anthropic in 2021 with other former employees of OpenAI, the company that developed ChatGPT. A prominent advocate of online safety, he is known for repeatedly warning about the dangers of unchecked AI development. He wrote that the world was substantially closer to real danger in 2026 than in 2023, when discussion of AI’s existential risks rose to the top of the political agenda.
He also referred to the controversy over pornographic deepfakes produced by Grok, the system developed by Elon Musk’s AI company, which flooded the social media platform X over Christmas and New Year. That included warnings that the chatbot was generating child sexual abuse material.
Amodei argued that some AI companies had shown alarming carelessness over the sexualisation of children in today’s models. That, he said, raised doubts about their willingness or ability to address the risks posed by autonomous systems in future.
Nobel-level capabilities
Anthropic’s chief executive believes powerful AI systems capable of autonomously building their own systems could be only one or two years away.
He defined “powerful AI” as a model more intelligent than a Nobel laureate in fields including biology, mathematics, engineering and writing. Such a system could give instructions to people or receive instructions from them. Although it would operate through a computer screen, it could control robots and even design them for its own use.
While acknowledging that powerful AI could be considerably further away than two years, Amodei said the technology’s recent rapid progress had to be taken seriously. If exponential progress continued, which was uncertain but now supported by a decade-long track record, he argued that AI could surpass humans at almost everything within a few years.
Last year, Amodei warned that AI could halve entry-level white-collar jobs and push overall unemployment to 20% within five years. In his essay, he cautioned that the economic gains from AI, including productivity improvements achieved by eliminating jobs, could become so large that nobody would want to apply the brakes.
That, Amodei said, was the trap: AI was such a powerful and attractive prize that human civilisation would find it extremely difficult to impose limits. Nevertheless, he remained optimistic. With decisive and careful action, he believed the risks could be overcome and the chances of success were good. An infinitely better world might lie on the other side, he argued, but humanity first had to recognise the scale of the challenge to civilisation itself.
How can you spot at a glance whether a text was written by AI?
