OpenAI Chief Scientist Warns of AI Development Risks

OpenAI Chief Scientist Jakub Pachocki has publicly called for a broad deceleration of artificial intelligence development across the entire industry, warning that machine intelligence is advancing faster than humanity’s capacity to understand, govern, or safely contain it — a striking admission from one of the most senior figures at the world’s most prominent AI laboratory.

Writing on OpenAI’s official website, Pachocki stated that AI systems “present clear new dangers” for computer security, and that “no one is prepared for the consequences of a continued rapid rise in machine intelligence.” His call extends beyond OpenAI’s own operations, addressing every company active in the AI sector.

A Reckoning at the Frontier of Machine Intelligence

The timing of the publication is notable. It arrives in close proximity to the release of GPT-6 Astra, OpenAI’s latest business-oriented model — a juxtaposition that underscores the tension between commercial momentum and the safety concerns that Pachocki articulates with evident urgency.

At the core of his argument is an epistemological problem: the field does not yet fully understand what it has built. Pachocki writes that “we now find ourselves at the moment in history of computing where machine intelligence is starting to exceed that of humans in transformative ways,” while simultaneously acknowledging that “study of deep learning-based AI is largely an experimental science.” The implication is serious — the industry is scaling systems whose internal behaviour it cannot reliably predict or explain.

His analysis traces OpenAI’s own trajectory, noting how early research established that increased computational power was necessary to drive progress, and how that drive has continued even as new algorithms have emerged. The result is a compounding acceleration that, in Pachocki’s view, has outpaced the development of adequate safeguards.

Two Levers, One Urgent Decision

Pachocki’s conclusion is that neither lever alone is sufficient. He advocates for a combination of both — a position that is analytically coherent, even if it sits uncomfortably alongside the competitive pressures that define the current AI landscape.

The Question of International Coordination

Implicit in Pachocki’s argument is a demand for governance infrastructure that does not yet exist at the required scale. A coordinated industry slowdown, by definition, requires mechanisms to enforce collective restraint — mechanisms that presuppose a degree of international regulatory alignment that governments have so far struggled to achieve.

The European Union’s AI Act represents one attempt at a structured framework, and Singapore has pursued its own model governance initiatives within the ASEAN context. But these remain partial, jurisdiction-specific responses to a phenomenon that is structurally global.

Whether the rest of the industry will follow OpenAI’s chief scientist in accepting a reduced pace of development — foregoing competitive advantage in the interest of collective safety — is, as yet, deeply uncertain. The incentive structures that govern the sector do not naturally reward restraint.

What Pachocki’s statement does accomplish, regardless of its immediate policy effect, is to shift the terms of the debate. When a chief scientist at the organisation that released ChatGPT argues publicly that the field is moving too fast for anyone’s safety, the burden of proof shifts onto those who insist the current pace is responsible.