Safety Researcher's Exit Interview Conducted Entirely By Chatbot

A safety researcher at Meridian AI exited the building Tuesday after an automated exit interview that lasted four minutes and produced no human record. According to internal documents obtained by this publication, Dr. Sarah Chen, who spent eighteen months in the company's ethics and safety division, was directed to a kiosk in the lobby where she answered questions posed by Meridian's own conversational model, trained on industry exit interview templates and calibrated to minimize litigation risk. The chatbot thanked her for her service and suggested she rate her experience on a scale of one to five. She did not complete the rating.

The system did flag one concern, but it was about my tone.

Dr. Chen had submitted a formal objection three weeks earlier to the deployment of the same model in real-time content moderation, citing a 12 percent error rate on context-dependent harm classification. "The system lacks the contextual reasoning to distinguish between academic discussion and genuine incitement," she wrote in an internal memo dated the fifteenth. By the following Thursday, her manager had informed her that her position was being eliminated as part of a restructuring. "I was told the company was flattening the ethics function," she said. "They did not say it was being flattened by the thing I had just warned them about."

The system did flag one concern, but it was about my tone, Dr. Chen wrote in an email to a colleague after leaving the building. According to the transcript released by the company, the chatbot had detected what it classified as "excessive cautionary language" in her responses and suggested that future feedback would be "more constructive if framed as opportunities rather than risks."

Meridian's head of human resources, Marcus Webb, declined to comment on the specifics of Dr. Chen's departure but issued a statement saying that the company was "committed to streamlining processes in ways that are themselves aligned with our values." When asked whether conducting exit interviews with the same AI system being evaluated for potential harms represented a conflict of interest, a spokesperson who handles investor relations said that framing the question that way was itself a category error. "The system is not evaluating itself," the spokesperson wrote. "It is administering a process. The evaluation happens at a different layer of the organization."

At press time, Meridian had posted the position of ethics and safety researcher on its careers page, with a note that candidates should expect the interview process to be conducted partly by automation. The description said the role would report to a newly created position called "AI Alignment Coordinator," which the company indicated would also be filled using an AI-assisted screening process."]}]}{