Anthropic survey reveals who fears AI the most

An Anthropic survey of 80,508 Claude users shows a clear tension: AI helps many people work more and better, but it also increases fears of job loss, especially among those at the start of their careers.

The study connects, for the first time, what people say they feel with the jobs where Claude already performs the most tasks. It is not a prediction of which jobs will disappear, but a snapshot of how users are experiencing the arrival of these tools.

The most exposed jobs generate the most concern

Approximately one in five participants expressed concern about job displacement. The concern was greater in occupations where Claude has a stronger presence, such as programming and other digital tasks.

For every 10-percentage-point increase in a job's exposure to Claude, concern about losing it increased by 1.3 percentage points. In addition, people in the 25% of occupations with the greatest exposure mentioned this fear three times more often than those in the 25% with the least exposure.

The difference also appears by professional experience. Workers at the beginning of their careers were much more likely to fear that AI would replace their jobs than senior professionals.

The result fits a specific concern: if a tool can perform part of a programmer's, analyst's or writer's work, companies may need fewer people for entry-level tasks. In some cases, users also said that AI had not taken work away from them, but had allowed their managers to assign them more difficult tasks.

Productivity increases, but not equally for everyone

Users who responded to the survey described significant improvements. On a scale from 1 to 7, the average productivity rating was 5.1, equivalent to feeling “substantially more productive.”

The figure has an important limitation: Claude users who chose to respond took part, and those who perceived benefits were probably more willing to do so. In addition, 42% did not provide enough information to evaluate their productivity, and 3% said they had experienced a negative or neutral effect.

The largest increases were concentrated at two extremes of the labor market:

• High-earning professionals, especially in management, programming and mathematics.
• People in low-wage jobs who use Claude to expand their capabilities or start their own projects.

For example, some users employed the tool to create online stores, music apps or customer responses. AI did not just help them finish a task sooner: in some cases, it helped them do things they did not know how to do before.

The main benefit is being able to do more

Among those who spoke explicitly about productivity, the most common benefit was expanding the scope of their work. 48% said they could now take on new tasks or capabilities. Another 40% highlighted that they completed activities they already performed more quickly.

The difference is simple. An accountant can use AI to complete in 15 minutes a process that previously took two hours. Someone without technical training can create an app. In both cases, the tool does more than speed up the work: it changes what that person can offer.

Most of those who identified a beneficiary said the gains went to them, through faster tasks, more free time or new responsibilities. However, 10% of that group said employers or clients were using AI to demand more work.

The generational gap also appears here: 80% of senior professionals who identified a benefit said the gain went to them, compared with 60% of workers in the early stages of their careers.

Those who speed up their work the most also fear more

The study found a U-shaped relationship between speed and concern about employment. Those who said AI slowed down their work also expressed considerable fear, especially in creative professions where the tool is seen as rigid or not very useful.

Among the remaining participants, the relationship was direct: the more AI sped up the work, the greater the fear of losing the job. It is a logical reaction. If a task that used to take hours can be completed in minutes, a worker may wonder how much value their position will have in a few years.

Anthropic warns that the results do not represent all workers. The survey included only users of personal Claude accounts who chose to respond, not a general sample of the population or most business users. In addition, the company inferred profession and experience from open-ended responses, so some classifications may be incorrect.

Even with these limitations, the study points to a reality that is already taking shape: AI can increase a person's capabilities while also making their position seem less secure. What needs to be watched now is who keeps that productivity gain: the worker, the company or the client.