1. Home
2. News
3. OpenAI

Helen Toner would vote to remove Altman again, saying the future can't hinge on one CEO

Three years after the vote, the former OpenAI board member says relying on any one person to get AI right is the mistake.

By Think Facility · November 14, 2026 · OpenAI AI safety Save

Helen Toner says she'd vote to fire Sam Altman again. She told Nilay Patel so on The Verge's Decoder podcast, in an interview published on November 14, 2026. In November 2023 she was one of the four OpenAI board members who voted to remove him. He came back as CEO and she resigned.

Patel asked whether she'd do it the same way. Her answer was short: "Oh, for sure." She added that the board knew going in that it would be miserable. Toner is now executive director of Georgetown University's Center for Security and Emerging Technology, and most of the hour was about what comes after trusting a CEO.

Why she doesn't think alignment gets solved

Toner pushed back on the way people in AI talk about solving things. On alignment, the work of getting AI to reliably do what people intend, she said: "I don’t know that alignment is a thing that can be solved." Even the narrow version, an AI that does what its user wants, isn't on track by her reading.

The labs' own deadlines are what worry her most. When the builders say the problem has to be fixed within 12 to 24 months, she called that terrifying. Her own guess is that they're probably wrong and there's more time, though she noted that very smart people disagree with her.

She was blunter about the fixes people reach for. She called better sandboxing a fallacy as an answer to autonomous cyberattacks, because the whole point of these systems is to give them more autonomy, permissions and ways to act in the real world.

What the labs could do now

Toner thinks there's a lot the companies could do today. People inside the frontier labs, she said, see hundreds of obvious safety fixes and don't have time for them. So they triage, doing two or three per fine-tuning run and two or three per launch.

Asked what a stop would even mean, she sketched one: stop training the next model, stop racing for commercial reasons, and spend that time working down the list. Patel noted that the CEOs have asked Washington for help with a slowdown, and that the White House answered with a voluntary accord that leans partly on the kind of board accountability that broke at OpenAI.

On artificial general intelligence, she won't say we have it. She does say the term has stopped being useful. In her picture AGI is a fuzzy cloud of definitions, and we're now inside the cloud, past some of them.

What it means for voters and investors

Patel and Toner spoke the day after the US midterm elections. She described a campaign of attack messages, opposition research and astroturfing on X against candidates who talked about AI safety or regulation, and said it really didn't work. Her reasons were that voters increasingly care about AI and that the issue is pretty bipartisan now. Public First made a similar claim after the vote, in its own count of the races it backed.

For people who'll own shares, her point is about boards. An independent committee of the board, she said, is close to toothless at a private company that decides things for itself. At a public company it has a pretty strong legal meaning. She said she's watching that with great interest as labs head for the stock market, Anthropic among them with its roadshow underway.

The Verge lists the episode as part two of a series on AI in real life.