SAN FRANCISCO, September 15 — Anthropic has spent five years arguing that the most dangerous technology in modern history should be built carefully, by people who take the danger seriously. The company’s own research suggests that same technology may soon take a large share of white-collar jobs. Both things are true at once, and that is the tension the AI safety firm has never fully resolved.
Founded in 2021 by former OpenAI researchers Dario Amodei and Daniela Amodei, Anthropic has built its public identity around responsible development of artificial intelligence. Its 2023 “Core Views on AI Safety” lays out the thesis plainly: advanced AI systems pose potentially catastrophic risks and must be developed with safety as the paramount concern.
The company operates as a public benefit corporation and publishes a Responsible Scaling Policy committing it to evaluate and mitigate risks as its models grow more capable. Its flagship product line, the Claude family of large language models, takes its name from Claude Shannon, the mathematician who founded information theory with his 1948 paper “A Mathematical Theory of Communication.”
The safety argument, in plain terms
Anthropic’s consistent position is that safety and speed of deployment are not in conflict. Building powerful AI under a safety-focused roof, the company argues, is preferable to leaving development to less cautious actors. Think of it like nuclear research: the people who understand the risk best are the ones you want holding the materials.
Critics, including some labor economists and technology ethicists, counter that this framing treats job displacement as an acceptable externality rather than a problem to be solved. The scale of that displacement is the subject of years of debate.
A 2023 Goldman Sachs report estimated that generative AI could expose the equivalent of 300 million full-time jobs globally to automation. A 2024 International Monetary Fund analysis found that nearly 40 percent of global employment is exposed to AI, with advanced economies facing greater exposure than emerging markets. In the United States, the Bureau of Labor Statistics tracks occupational change but does not yet publish a dedicated AI displacement series.
Anthropic’s own leadership has not dodged the subject. In a 2024 essay, CEO Dario Amodei wrote that AI could eliminate a large share of entry-level white-collar jobs within a few years, while arguing the transition can be managed if society invests in retraining and new forms of economic support.
The company has funded research into those labor market effects, including the Anthropic Economic Index, launched in 2025, which uses anonymized usage data to map how AI is being adopted across occupations and tasks.
What workers already won
The debate is not abstract. In 2023, the Writers Guild of America strike, which lasted 148 days from May 2 to September 27, secured contract language governing the use of generative AI in film and television writing. The following year, the Screen Actors Guild-American Federation of Television and Radio Artists strike — which ran from July 14 to November 9, 2023 — won protections around digital replicas of performers.
Those disputes put a human face on the question of who bears the cost when AI systems are trained on and then replace human labor. Anthropic’s competitors face the same scrutiny.
OpenAI, Google DeepMind, Meta and others have all published safety frameworks while racing to deploy increasingly capable models. The policy landscape is tightening around them: the European Union’s AI Act entered into force on August 1, 2024, and the United States’ executive order on AI, signed in October 2023, directed federal agencies to assess AI risks, though its implementation status has shifted with subsequent administrations. For readers, the practical upshot is this.
The key question is not whether Anthropic is sincere about safety. It is whether a company can credibly promise to manage the risks of a technology whose economic logic depends on rapid adoption.
The company’s founders have said they believe the answer is yes. The workers whose jobs are most exposed have yet to be asked. Watch next for the Anthropic Economic Index’s occupation-level data, and for whether retraining and economic-support proposals move from essays into actual policy.
