OpenAI Researcher Warns Pacing AI Development Won’t Prevent Existential Risks

Daniel Selsam, a senior researcher at OpenAI with nearly five years on the company’s model-training teams, issued a public statement on Monday expressing “extremely concerned” about the rapid advancement of AI models. Selsam did not define what “engineering” models would entail or propose concrete alternatives, noting he has no ready answers.

Business Insider

The phrase “pacing the frontier” has been promoted by OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei as a strategy to slow AI progress without halting development entirely. Both leaders have advocated for embedding independent safety auditors—“embedded evaluators”—who could review training and deployment workflows and publish risk assessments.

Business Insider

Altman reaffirmed his support for the pacing approach in a recent X post, emphasizing that slowing development does not mean stopping it and that safety interventions carry significant expense. He also highlighted plans to work with independent evaluators to increase transparency around model risks. OpenAI’s public communications did not include a direct response to Selsam’s concerns, and the company declined interview requests.

Business Insider

• “But it should be slower than it otherwise could be; interventions like safety cases and monitoring have significant costs,” — Sam Altman, like openai CEO

Business Insider

Selsam’s warning underscores a growing tension within the AI community between accelerating capabilities and managing existential threats. While senior executives continue to promote moderated progress and external audits, the researcher’s critique suggests that pacing alone may be insufficient to address deeper alignment challenges. The debate signals that policymakers, industry leaders, and safety researchers will need to explore additional mitigation strategies beyond merely adjusting the speed of development.

Business Insider