Employment Gap Widens to 19% for Young Workers in AI-Exposed Occupations

01Stanford study update: Employment gap widens to 19% for workers aged 22 to 25 in highly AI-exposed occupations

Young workers in occupations considered highly exposed to artificial intelligence are falling further behind their peers in less exposed fields, according to an updated study by Stanford University economists. The research focuses on employees aged 22 to 25, a group that includes many entry-level workers, rather than claiming that employment has declined by the same amount for all young people or all junior positions.

The August 2026 edition of “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence” updates and revises a paper published last year. With fresh data and refined statistical methods, the researchers found that employment levels for 22- to 25-year-olds in the most AI-exposed occupations were 19% below those of people the same age working in fields less exposed to AI disruption.

That gap was 13% in last year’s study. Its increase to 19% indicates that the employment divergence identified earlier has continued and widened as more data became available.

The researchers based their analysis on a large subset of anonymized, high-frequency payroll records aggregated by human-resources management company ADP. They assessed each occupation’s exposure to AI using two measures: an existing gauge of AI’s potential labor-market impact and the Anthropic Economic Index, which tracks how workers in different occupations actually use Anthropic’s Claude model in their everyday work.

Combining those measures allowed the researchers to distinguish occupations expected to face greater AI disruption from fields with lower exposure. The comparison was then made within the same age group, rather than between younger and older workers. The reported 19% therefore describes the difference between 22- to 25-year-olds across those occupational categories, not a 19% decline across the entire youth workforce.

The update suggests that younger, entry-level employees in some fields are experiencing substantial employment losses, while older workers have so far appeared largely unaffected. But the available account does not establish that AI alone caused the entire gap. It does not provide the specific occupations, sample size, observation period or complete set of revised statistical controls needed to separate AI’s contribution from every other labor-market influence.

The study’s strongest finding is consequently the widening association: the occupations rated as most exposed to AI now show a larger employment shortfall among workers aged 22 to 25 than they did in last year’s analysis. How much of that change is directly attributable to AI remains unresolved.

02OpenAI releases first Jalapeño benchmarks, claims up to 1.9 times the inference work per watt of Nvidia comparison systems

OpenAI has published the first performance results for Jalapeño, its custom chip for AI inference—the process of running a trained model to answer requests or operate an agent. Introduced in June and made with Broadcom, the application-specific integrated circuit is intended to increase throughput while reducing response latency.

The company said Tuesday that Jalapeño delivered more work per watt and faster responses than the best Nvidia-based results then recorded on the InferenceX benchmarking platform. OpenAI plans to move from testing to deployment in small volumes by the end of this year, followed by a production ramp during 2027.

The benchmark covered three models: GPT-OSS 120B, DeepSeek R1 and Kimi K2.5 1T. Jalapeño was compared with InferenceX’s best available results at the time, which used Nvidia GB200 or GB300 systems. OpenAI said its chip completed between 1.5 and 1.9 times as much AI work per watt across those models.

OpenAI also said Jalapeño’s end-to-end latency was between 1.7 and 3.6 times lower than that of the comparison systems across the three models. Richard Ho, OpenAI’s vice president of hardware, said the results combine lower latency with higher throughput, two measures that AI systems commonly have to balance. According to Ho, the practical benefits could include faster answers, more responsive agents and more reliable access as demand increases.

The reported range applies specifically to the three tested models and the InferenceX comparison results, rather than to every inference workload. The sources do not provide the full test configurations, deployment costs or independently reproduced results.

OpenAI is also not preparing to replace its entire existing chip fleet with Jalapeño. Ho said the company’s broader computing strategy will continue to include partners such as Nvidia, while OpenAI develops second- and third-generation versions of its own chip.

The company has not disclosed how many Jalapeño chips it expects to deploy this year or during the 2027 expansion.

03Stability AI Raises $76 Million Series B With Universal, Sony, Warner and EA Participating

Stability AI, the generative AI company behind the Stable Diffusion image model, has raised $76 million in Series B funding, bringing its total fundraising to $232 million. The company also develops models for generating music and video, and has spent the past year working to introduce generative tools into entertainment production.

The round deepens that push by making several entertainment companies both investors and development partners. Stability AI said part of the new capital came from Universal Music Group, Sony Music Group, Warner Music Group and gaming company Electronic Arts. AMD Ventures and Pacific Alliance Ventures also participated.

Stability AI plans to use the money to expand its suite of “creative production” products and its professional services business. That combination suggests the company is pursuing both software tools and hands-on work helping professional customers adopt them, although it did not disclose how much funding would go to either area.

Some of the investors were already helping shape those products. Stability AI formed partnerships with Universal Music and EA in October 2025, followed by Warner Music in November. Under those arrangements, the companies participate in co-developing Stability AI’s tools rather than merely licensing the content they produce. The source did not describe a comparable existing partnership with Sony Music.

The dual role gives the entertainment companies a direct stake in a supplier whose technology they are helping adapt to their workflows. Stability AI, in turn, gains capital alongside access to potential customers and collaborators in music and games. The company did not disclose its valuation, individual investment amounts, transaction terms or whether any participant received governance rights.

The financing arrives while Stability AI still faces legal uncertainty over the material used to train image-generation models. The company largely prevailed in a United Kingdom copyright lawsuit brought by Getty Images, which alleged that its images had been used in training without respecting its intellectual-property rights. A similar Getty case remains pending in the United States, meaning the new investment does not resolve the company’s broader copyright exposure.