# OpenAI's policy chief says frontier lab insiders are worried about AI, too > Source: > Published: 2026-09-02 13:39:37+00:00 # OpenAI's policy chief says frontier lab insiders are worried about AI, too **Dean Ball, hired in July to shape frontier policy, says he will keep issuing candid warnings even when they complicate OpenAI's message.** By [Ryan Merket](/author/ryan-merket) · Published · Updated Primary source: [X](https://x.com/deanwball) ## Why it matters OpenAI has given a senior policy official room to voice fears its marketing cannot smooth over. His warnings will also test whether the lab acts on concerns raised from inside its own leadership. [Dean W. Ball (@deanwball)](https://x.com/deanwball), OpenAI's head of strategic futures, said on September 1st that public concern about artificial intelligence is often justified and that his job does not require him to make the technology feel safe. "The public, very often, is right to be concerned. I am concerned," Ball wrote on X. He added that most of the people he knows at frontier AI labs share concerns about where the technology is heading. The statement came less than two months after Ball joined OpenAI and draws a clear boundary between his role and the company's marketing operation. Ball said he was not hired to repair AI's image or reassure people about using tokens. He described a technology that has enriched his own life and could benefit many others, while producing changes that he expects to be "extraordinarily fraught." Ball arrived at OpenAI with a background in policy rather than product promotion. Before starting at the lab on July 6th, he served as a senior policy adviser for AI and emerging technology in the White House Office of Science and Technology Policy and as a strategic adviser for AI at the National Science Foundation, according to his [Foundation for American Innovation biography](https://www.thefai.org/profile/dean-ball). The foundation says he also co-authored the White House AI Action Plan and remains one of its non-resident senior fellows. OpenAI created the Strategic Futures team to examine frontier AI policy, governance and the institutional changes that advanced models could force. When Ball's appointment was announced in June, OpenAI Chief Strategy Officer Jason Kwon told [Axios](https://www.axios.com/2026/06/18/dean-ball-openai) that he expected Ball to pressure-test the lab's thinking and that the two would not always agree. Ball's public candor therefore matches the role OpenAI said it was hiring him to perform. It also creates a recurring communications problem for the lab: warnings from a senior OpenAI official carry the company's name even when Ball labels them as personal views. OpenAI formalized that separation on August 20th when Ball launched [Intelligence Age, the Strategic Futures team's blog](https://openai.com/index/introducing-intelligence-age/). Each post carries a disclaimer stating that its arguments represent the author rather than OpenAI's organizational position. The team identified concentration of power as its central concern and said AI could restructure companies, governments and civil society as profoundly as industrial technology reshaped the modern corporation. Ball's September 1st post broadens that work into a defense of public warning itself. He said some expected developments cause him fear and sadness in his private life, and that he considers it a duty to describe them. He rejected pressure to soften that message, writing that he was not there to "sing you some lullaby." The dispute arrives as AI safety work becomes increasingly automated and quantitative. The [Foundation's Edge analysis](https://seldondance.substack.com/p/the-industrialization-of-ai-safety) circulated alongside the discussion was originally dated December 28th, 2025, rather than September 2026. It examined Bloom, an automated behavioral evaluation framework that Anthropic had released nine days earlier. According to [Anthropic's December 19th announcement](https://www.anthropic.com/research/bloom?mibextid=UyTHkb), Bloom uses AI agents to design scenarios, conduct multi-turn interactions and score model behavior. Anthropic used it to build evaluations for delusional sycophancy, long-horizon sabotage, self-preservation and self-preferential bias across 16 frontier models. Anthropic said those evaluations took days to develop, while conventional evaluation work can become obsolete as models improve or benchmark material enters training data. Bloom also captures the harder problem beneath Ball's warning. Frontier labs increasingly use models to test other models, allowing safety evaluations to run at a scale human researchers cannot match. The resulting measurements still depend on the behavior selected by researchers, the scenarios generated by evaluation agents and the reliability of the models judging the answers. Automation expands the testing capacity without eliminating human judgment about which risks matter. Ball's remit is broader than model evaluation. His team is focused on the distribution of power, individual rights and the governance of AI systems capable of acting in the world. His post places public candor inside that mandate: society cannot prepare for disruptive systems if the people closest to them discuss only benefits. OpenAI now has a senior policy official openly saying that reassurance is outside his job description. The lab benefits from the credibility of employing an internal critic, while Ball's warnings raise a standard for OpenAI's conduct. Each concern he voices invites the same follow-up: what the company building the technology plans to do about it.