What Anthropic did

On March 5, Anthropic, the developer of the Claude AI, published a new metric — observed exposure. The aim of the measure is to quantify what portion of working time across different occupations is already covered by real-world scenarios of automation or significant task acceleration enabled by AI.

"The metric shows what share of working time and tasks in an occupation already fall under practical automation scenarios."

— Anthropic (press release, March 5)

Who Anthropic identified as most vulnerable

Anthropic released a top‑10 list of occupations with the highest exposure. This is not a verdict, but a risk marker for jobs and employment policy:

• Programmers — 74.5%
• Customer support specialists — 70.1%
• Data entry operators — 67.1%
• Medical records specialists — 66.7%
• Marketers and market analysts — 64.8%
• Sales managers — 62.8%
• Financial and investment analysts — 57.2%
• Software testers — 51.9%
• Cybersecurity specialists — 48.6%
• Technical support specialists — 46.8%

The percentage does not indicate inevitable layoffs, but the portion of work that modern models or automated workflows can already perform.

Context: other studies and deployment realities

Anthropic is not alone in such assessments: Microsoft Research last year compiled its own list of occupations with a large overlap with generative AI. At the same time, researchers at the Haas School of Business (UC Berkeley) found that deploying generative AI often does not reduce workloads — it changes them and sometimes even increases them.

"Some industry leaders predict rapid changes for office work; other studies point to growing workloads due to increased productivity."

— Mustafa Suleiman / Microsoft and UC Berkeley analysts (comparative context)

What this means for Ukraine

First, a slowdown in hiring young people (ages 22–25) in vulnerable occupations is already being recorded — this is a signal for education programs and employers. Second, some workers are in low‑exposure groups (cooks, mechanics, rescuers, bartenders, etc.): these occupations currently provide local employment resilience.

For Ukraine this suggests three practical conclusions: 1) invest in retraining and digital skills where AI augments rather than replaces work; 2) strengthen the security sector — both human and technical — because demand for cybersecurity specialists is changing, not disappearing; 3) leverage Ukraine’s IT legacy to create high value‑added jobs where exposure can become an advantage.

Expert view and risks

Experts agree: there is a gap between what AI can theoretically do and how it is actually applied in business. That is why metrics like observed exposure are useful — they help policymakers and employers respond proactively, rather than after unemployment rises.

Conclusion

Anthropic’s data send a clear signal: changes are coming gradually and unevenly. The Ukrainian labor market needs a combination of education, innovation incentives, and social support for those most affected by the technology. Whether predictive figures will be turned into state and business policy is a question on which the future workforce balance depends.