OpenAI 900 Career Research Debunks AI Unemployment Panic: 18% High-Risk Groups Have the Most Stable Jobs? What's Going On

OpenAI releases new research on over 900 occupations, with conclusions more optimistic than most expected. Workers in high-automation-risk jobs such as data entry, bookkeeping, and customer service (accounting for 18% of total employment) are experiencing a slower rise in unemployment rates than those in low-risk jobs.
(Background: DWF Ventures reports that AI accounts for 19% of DeFi trading volume but performs at only one-fifth of human capability in complex transactions)
(Additional context: Perplexity Personal Computer officially launched: AI now manages Mac locally, $200 monthly fee available to Max users)

Table of Contents

Toggle

• Are high-risk groups actually holding up the best?
• Demand growth offsets layoffs
• High-risk groups have only used 1/4 of AI’s theoretical capabilities; real impact has yet to begin

Eighteen percent of jobs are classified by OpenAI as high-automation-risk, yet this group did not lose jobs first and actually shows the slowest increase in unemployment? This is the key insight from OpenAI’s report on 900 occupations released on the 16th, and it’s also the part that leaves people stunned after reading. (Or is it a deliberately biased AI-favorable perspective?)

Data entry clerks, bookkeepers, and customer service reps—these three job categories have long been cited as being replaced by AI. But OpenAI’s report shows they are already handling about three times the task volume of other occupations, and their unemployment rate is rising more slowly than those in “low-risk” jobs.

Are high-risk groups actually holding up the best?

OpenAI’s 2023 report sparked widespread discussion, with critics pointing out that it only measured “exposure” without tracking actual usage rates and unemployment data. The new report on April 16th adds these two dimensions, and the conclusions are almost reversed.

The four categories are outlined as follows:

• 18% are high-automation-risk groups (data entry, bookkeeping, customer service)
• 46% are minimally affected groups (teachers, domestic workers, etc.)
• 24% are downsized but still require human oversight
• 12% are expansion groups (software development, etc.)

But “high automation risk” does not mean “job will be taken.” OpenAI’s own research provides a counterexample. Not only did high-risk groups not exit first, but they are currently among the groups with the highest AI usage density.

Demand growth offsets layoffs

The report argues that when AI makes a service cheaper and faster, the overall market demand for that service often grows exponentially, offsetting the layoffs caused by efficiency gains.

For example, in programming: the cost of coding decreases, but demand for software skyrockets, leading to an expansion of software development positions. This is the reason why 12% of the “expansion” group is growing.

In simple terms: demand growth consumes the efficiency gains, which is the core message of this report.

High-risk groups have only used 1/4 of AI’s theoretical capabilities; real impact has yet to begin

But buffer zones are not forever safe. The report points out that current workers in high-automation-risk jobs are using less than a quarter of AI’s theoretical capabilities. In other words, the real stress test has not yet arrived.

DWF Ventures just reported on April 17th that AI accounts for 19% of DeFi trading volume but still performs at only one-fifth of human capability in complex transactions. More work, less replacement. OpenAI’s report echoes this scenario on a larger stage of 900 occupations, signaling that AI’s penetration is deeper than most imagine, but the extent of replacement remains limited by capability boundaries.

If usage continues to increase or AI’s capability boundaries expand rapidly, the current balance supported by consumer elasticity could be broken. The buffer for high-risk groups is real, but it relies on demand growth continuously catching up with efficiency improvements—an assumption that is not guaranteed to hold forever.