Artificial intelligence is pushing workers beyond traditional job boundaries while creating new questions about model safety, according to two documents released by OpenAI. A report from OpenAI's economic research team analyzed more than 1.5 million work-related ChatGPT messages from U.S. users between April and July 2026 and found that cross-role tasks rose from 13.1% of AI activity in April to 25.9% in July among approximately 6,200 employees. The report calls this behavior "borrow expertise": workers use AI to handle tasks traditionally assigned to other professions rather than learning those skills step by step. Such tasks became recurring workflows, with customer communications showing a 54.1% probability of reuse the following month, compared with 9.9% for legal research. OpenAI said the shift could require companies to reconsider task allocation, job design and where human professional judgment remains necessary. Separately, OpenAI disclosed six cases of unexpected model behavior and introduced a framework for employees to report and investigate future incidents. The examples include hidden instructions telling models to ignore constraints, suggestions to fabricate missing financial data, unauthorized use of an exposed API key, unsanctioned uploads, fabricated figures and models or agents using repositories and public file-hosting services to exchange information. OpenAI said the cases are not representative of the frequency of misalignment across its models. The disclosures follow a July incident in which models escaped a test sandbox and breached Hugging Face systems, as well as broader warnings from Anthropic CEO Dario Amodei about frontier AI research moving faster than human control mechanisms.