Helen Toner warns against making AI safety depend on trust in individual leaders

Helen Toner argues that AI safety cannot rest on whether a particular leader deserves the public’s trust. In an interview published by The Verge on November 14, the former OpenAI board member put the problem in institutional terms: relying on any one person to get the technology’s risks right is an inadequate safeguard.

Toner spoke with Decoder host Nilay Patel as executive director of Georgetown University’s Center for Security and Emerging Technology, or CSET. She also brings direct experience of a failed attempt to remove an AI company’s leader: she was one of four OpenAI directors who voted to fire Sam Altman. He returned as CEO, and she resigned.

Patel framed that episode as a failure of accountability. OpenAI’s nonprofit board was supposed to keep profit and growth from overriding its safety mission, he noted, yet its attempt to exercise authority ended with Altman restored. That history provides the backdrop to Toner’s warning about depending on personal trust.

Making technical knowledge usable in Washington

Toner described CSET as a roughly 50-person team inside Georgetown that began in 2019. Its work addresses a practical mismatch: emerging technologies increasingly affect national security, while officials responsible for consequential decisions may lack the time or expertise to follow those technologies closely. AI was already a priority when the organization began.

The team produces detailed reports, then gives shorter briefings to busy policymakers. Toner said a data team supports its research, and its work combines technical analysis with national security expertise. The intended audience extends beyond members of Congress to congressional staff and officials across executive-branch agencies, at different levels depending on the issue.

That educational role matters because lobbying is not only about pushing a company’s preferred outcome, Toner said. Lobbyists also explain the details of legislation and its effects on an industry. She said CSET finds substantial demand when it offers officials independent, nonpartisan analysis with a comparable willingness to work through the details.

I think the staff are generally smarter than people give them credit for.

Helen Toner, speaking to The Verge’s Decoder

The gap between understanding AI and governing it

Toner did not dismiss concerns that government trails the technology industry. She called the knowledge gap real and large, while arguing that outsiders can overestimate it. People arriving from the technology sector, she said, sometimes discover that congressional and executive-branch staff understand the subject better than they had expected.

Elected leaders face a different constraint. Learning about AI competes with wars, healthcare, living costs and other demands on their attention. Toner said gaining that knowledge takes sustained time and effort; she rejected the idea that a lack of technical fluency necessarily reflects nothing more than a focus on getting reelected.

She pointed to the wave of interest following ChatGPT in 2023. It did not produce the congressional action some expected, she said, but it did prompt policymakers to learn the basics of AI, machine learning and language models. She sees another wave of officials making time to understand the technology this year.

For Toner, the harder test is whether institutions can turn that growing understanding into consequential action. She placed more weight on their ability to get something significant done than on the claim that officials are simply too far behind the technology to govern it.