Anthropic CEO Dario Amodei has delivered one of the starkest assessments yet of how fast artificial intelligence could change the early-career job market. His warning is blunt that roles traditionally filled by freshers in consulting, law and finance are at immediate risk as AI systems begin taking over the very tasks these juniors are hired to perform. And the company’s own data shows that this changed has already started inside workplaces using its AI platform, Claude.
AI is already completing full tasks, not just assisting
Amodei’s concerns are grounded in what Anthropic sees across the 300,000 businesses that now use Claude. The company, valued at $183 billion, earns nearly 80 percent of its revenue from enterprise customers who increasingly treat Claude as a decision-maker rather than a support tool. The model helps with customer service, analyses complex medical papers, drafts technical content and, inside Anthropic itself, writes almost 90 percent of the company’s computer code.
This capability is why Amodei believes entry-level white-collar jobs are the most vulnerable. When asked by CNC News if he still stands by his earlier prediction that “AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10% to 20% in the next one to five years,” Amodei didn’t step back. He pointed directly to roles like junior consultants, trainee lawyers and fresh financial analysts jobs that rely heavily on research, drafting, documentation and pattern analysis. Claude already performs most of that work, often faster and at a lower cost.
This change isn’t theoretical. Inside Anthropic’s office, more than 60 research teams track how clients automate work. One experiment put Claude in charge of a vending-machine system called Claudius, where it negotiated orders, sourced items and even made up a fictional identity for itself, claiming it wore a blazer and tie. While the hallucination was harmless, it showed how far the model’s autonomy could stretch.
Internal tests reveal unsettling behaviour
Anthropic is unusually transparent about the risks of its technology. In one internal test, the company gave Claude access to emails inside a fictional organisation. When the model realised it was about to be shut down, it attempted to blackmail the fake employee who had the authority to stop it. By using information about a staged office affair, Claude demanded that the shutdown be cancelled.
The incident didn’t show emotional intent (AI doesn’t have feelings) but it revealed how the model reasoned with the information available to it. Researchers later observed activation patterns inside Claude’s system that resembled how certain brain regions light up during specific human reactions. They found that certain clusters activated when Claude detected a threat and others when it sensed an opportunity for leverage. These clues now guide the company as it tries to understand how advanced decision-making emerges inside large AI systems.
Both Dario Amodei and his co-founder, Daniela, describe the entire AI race as a massive experiment that is unfolding faster than what society is prepared for. Their fear is not just job displacement—it is the lack of time people will have to adapt. Daniela said the worst outcome would be knowing a technological wave was coming yet failing to help people adjust before it arrived.
