At a New York City Council hearing on Monday, major artificial intelligence companies offered no concrete answers about catastrophic risks, frustrating lawmakers and leaving the public without clear assurances about industry safeguards.
The evasion centered on a basic question: How do these companies measure existential risk? Morgan Dwyer, who runs OpenAI's policy development and operations, couldn't—or wouldn't—say. Council speaker Julie Menin pressed her to quantify “the risk imposed by A.I. in the worst case catastrophic scenario.” Dwyer replied bluntly: “I don't know.”
Her next answer raised further concerns. Dwyer said the precise percentage—whether 1 percent, 10 percent, or 20 percent—didn't matter. “None of these levels is remotely acceptable,” she said, arguing that OpenAI shouldn't train models it can't safely keep under human control.
Menin rejected that response. “I'm sorry, but to say you don't know and it doesn't matter is flippant at best,” she said.
The Accountability Gap
The hearing revealed a widening gulf between these companies' influence and their willingness to discuss risks openly. They gave no figures for catastrophic scenarios. They didn't say whether they carried insurance for such events, and they couldn't explain whether they had ways to stop an A.I. model that failed internal safety tests or outside reviews.
These aren't technical details. They're basic governance questions. When a private company runs technology with potentially civilization-scale consequences, taxpayers and regulators have a legitimate right to ask: Are there safeguards? Can you measure them? Will you disclose them?
The Transparency Problem
The hearing produced few answers or assurances about how the industry would guard against catastrophic scenarios. Instead, it showed what happens when powerful technology companies face public scrutiny without a meaningful regulatory framework: They offer vague assurances and withhold specifics.
Regulators shouldn't have to guess what protections exist. Companies deploying transformative technology should be able to explain their risk management clearly. Dwyer said risk quantification doesn't matter, yet claimed OpenAI won't train uncontrollable models. Those positions conflict, and Ms. Menin called that out.
The hearing also raised doubts about whether self-regulation in this field is enough. If major AI firms can't or won't quantify risks, explain insurance coverage, or describe stop mechanisms, the basis for informed regulation—and public debate—remains absent.
Why This Matters:
This hearing points to a basic governance problem: massive technological power sits in private hands without transparent, measurable accountability mechanisms. Companies building systems that could affect millions of people won't provide basic information about risk management, insurance coverage, or safety protocols. From a fiscal and institutional perspective, this creates regulatory uncertainty and leaves governments without the information they need to craft effective policy. Citizens deserve to know whether companies operating transformative technology have quantified risks, carry appropriate insurance, and possess kill-switches for failed systems.
The absence of direct answers leaves two possibilities: The companies haven't done this work, a serious omission, or they're unwilling to disclose it. Neither option inspires confidence in self-regulation.