Rosie Campbell Left OpenAI—Now She Is Eleos’s Managing Director
Rosie Campbell remains outside OpenAI and now serves as managing director of Eleos AI Research. The current Eleos team page identifies her as an AI-governance researcher with previous experience on frontier-policy issues at OpenAI; it also identifies Eleos as the operating name of a nonprofit institute.
Her resignation occurred in late 2024, not in 2026. In Campbell’s December 1, 2024 account, she placed her departure after almost three and a half years at OpenAI in the context of Miles Brundage’s exit and the dissolution of the AGI Readiness team. She also expressed concern about organizational changes, the loss of colleagues who had influenced the company’s culture and whether existing safety practices would remain adequate for substantially more capable systems.
The resignation was an institutional warning, not evidence of a safety incident
Campbell did not identify a concealed accident, a particular model failure or proof that OpenAI had developed uncontrollable artificial general intelligence. Her concern was about the institutional conditions for anticipating risks: which questions could be pursued internally, how much weight the company gave them and whether its approach could keep pace with more capable systems.
That boundary matters because a senior researcher’s resignation can easily be treated as proof of claims the researcher did not make. The departure supports a narrower conclusion: Campbell believed that the dedicated setting for her work on advanced-AI policy and readiness had disappeared, and that she could pursue that work more effectively outside the company.
Her message also distinguished the creation of advanced AI from the obligation to make its effects beneficial. The warning was therefore not an argument against developing the technology in all circumstances; it challenged the assumption that current governance and evaluation methods would necessarily remain sufficient as capabilities increased.
What disappeared with the AGI Readiness team
Campbell’s work crossed boundaries that conventional product or security teams do not always cover. Her portfolio included dangerous-capability evaluations, governance for agentic systems, digital sentience and policy questions associated with potentially transformative AI.
Those subjects connect technical evidence to decisions about deployment and institutional responsibility. A capability evaluation may reveal what a model can do, for example, but governance determines who reviews that evidence, what threshold triggers intervention and whether the result can alter a release plan.
The end of one team does not establish that OpenAI abandoned every related project or eliminated all safety work. It does show that a researcher engaged with these questions no longer saw an appropriate internal home for the particular combination of frontier policy, readiness and exploratory research she wanted to pursue.
OpenAI’s newer framework addresses a different layer of the problem
OpenAI has since made its formal risk-management structure more explicit. Its Frontier Governance Framework published May 28, 2026 places the Preparedness Framework at the foundation of its approach and covers cyber offense, chemical and biological threats, harmful manipulation, loss of control, model reporting, incident response and external expert input.
The document is a meaningful expansion of the company’s public governance record, but it does not directly reverse the organizational change behind Campbell’s departure. OpenAI has not publicly presented the framework as a restoration of the former AGI Readiness team with the same remit, personnel or position inside the company.
A written framework and an internal research culture also answer different questions. Formal procedures can define risk categories, assessments and escalation routes; they do not, on their own, reveal how much authority dissenting researchers hold or whether an unfavorable assessment will change a commercial decision.
Campbell’s Eleos role extends part of her earlier agenda
Campbell’s move to Eleos keeps her within AI governance while placing her in an organization focused on AI consciousness, welfare, agency and moral status. These subjects differ from familiar misuse risks such as cyberattacks, yet they share a forward-looking governance problem: institutions may need to prepare for uncertain consequences before those consequences become immediate product issues.
The move also illustrates what independent research can offer. An external organization can pursue neglected questions without tying each project to a laboratory’s release schedule, although it generally lacks the privileged access to unreleased models, internal evaluations and decision-makers available to company researchers.
That trade-off helps explain why the departure remains relevant beyond a personnel change. Frontier laboratories possess the systems and evidence needed to examine emerging risks closely, but the same organizations also build and commercialize the technology that safety research may constrain.
Why the departure still matters
The subsequent developments do not resolve Campbell’s original concern. She has moved into nonprofit leadership, and OpenAI has created a broader public account of frontier-risk governance, but neither fact demonstrates whether internal critics now have more influence over consequential decisions.
The durable issue is institutional authority rather than the existence of safety language or evaluation programs. Research can identify a risk, and a framework can describe how that risk should be handled, but effective governance also depends on whether the people raising difficult findings have a protected route to affect strategy and deployment.
Campbell’s exit remains a documented example of that tension. An experienced policy researcher concluded that important work on advanced-AI readiness no longer had a suitable internal setting; her current role shows that she continued the broader agenda elsewhere, while OpenAI’s later framework shows how the company now publicly organizes severe-risk management.
Also read:
Subscribe to our newsletter
Get the latest Web3, AI, and crypto news delivered straight to your inbox.