OpenAI

OpenAI is an artificial intelligence developer. OpenAI, Inc. was incorporated as a Delaware nonprofit corporation in 2015, with the mission of “ensuring that artificial general intelligence benefits all of humanity,” according to Delaware’s attorney general. On October 28, 2025, after a review of more than a year, the attorney general issued a statement of no objection to its recapitalization: the nonprofit keeps control of a newly formed public benefit corporation (PBC), including the sole power to appoint and remove its directors; the PBC’s mission is identical to the nonprofit’s; and a nonprofit board committee on safety and security can require mitigations up to halting the release of a model.¹ California’s attorney general said the same day that the concessions the attorney general's office secured included a commitment that OpenAI would remain in California.²

MetaRoles covers OpenAI through the deployments its models power, where an employer runs the work. Klarna said on February 27, 2024 that its customer-service AI assistant “powered by OpenAI” had held 2.3 million conversations in its first month, two-thirds of its service chats, and was doing work it equated to 700 full-time agents.³ In its annual report for 2025, Klarna says it formed a partnership with OpenAI in 2023, that the assistant handled 80% of its customer service chats in 2025 and did the work equivalent of over 850 full-time agents, and that it delivered approximately $59 million in cost savings; all are company-stated figures, examined in Klarna’s AI customer service claims.⁴ Lowe’s launched Mylow Companion, a generative AI tool for associates’ sales-floor devices, to all associates across more than 1,700 stores on May 5, 2025, and says it and the customer-facing Mylow advisor were both developed in collaboration with OpenAI; that case appears in AI in-store sales assistants.⁵

OpenAI researchers also wrote one of the exposure studies MetaRoles sets beside its own task counts. The “GPTs are GPTs” working paper assessed occupations against large language model capabilities, combining human expertise with GPT-4 classifications, and estimated that around 80% of the US workforce could have at least 10% of their work tasks affected; its authors say they make no predictions about adoption timelines.⁶ The 2024 version in Science lists two of its four authors at OpenAI in San Francisco.⁷ MetaRoles treats that figure as exposure, not job loss, in how MetaRoles measures AI in roles.