Hermes Vector

AI & SI

OpenAI board member says AI CEOs cannot be trusted

Helen Toner, a former OpenAI board member, says AI safety cannot depend on any one person being trustworthy and warns that current accountability mechanisms have failed.

By Hermes Vector

Also in: Español · Français · हिन्दी

Checked against 3 sourcesThe Vergefinance.biggo.comtechtimes.vn
Illustration for: OpenAI board member says AI CEOs cannot be trusted

In brief

  • Helen Toner says AI CEOs still cannot be trusted and safety must not depend on any one person.
  • Toner was one of four OpenAI board members who voted to fire Sam Altman in November 2023.
  • She says AI safety work is being triaged and is probably not solvable on the labs' 12-to-24-month timeline.
  • She favors a higher safety bar before model training or release rather than a literal pause.

OpenAI board member says AI CEOs cannot be trusted

Helen Toner, a former OpenAI board member and executive director of Georgetown University's Center for Security and Emerging Technology, says AI CEOs still cannot be trusted and that AI safety should not depend on any one person being trustworthy.

The OpenAI board episode

Toner served on OpenAI's nonprofit board when four board members voted to fire Sam Altman. The episode, sometimes called "the blip," ended with Altman returning as CEO, Toner resigning, and a new board of directors. OpenAI has since restructured entirely.

The nonprofit board existed to make sure the company did not betray its mission to build safe AI by prioritizing profit and growth over safety. When the board tried to act, that accountability mechanism broke.

Toner said she would relive the November 2023 decision to remove Altman, adding that they went into it knowing it was going to suck. Asked whether Altman is trustworthy, she said her stance is "pretty clear" and pivoted to structure: "If our hopes for the future are pinned on any one individual being trustworthy, then we're pretty screwed."

AI safety and regulation

Toner said the core AI safety problems — alignment, control, and misuse — may be solvable in principle but probably not on the 12-to-24-month timeline the frontier labs say they are working to. She described the binding constraint as time, not feasibility.

She favors setting a higher bar for confidence in risk management before training or releasing a model, rather than a literal pause. She also supports decelerating vertical growth in raw capability while accelerating horizontal growth in fine-tuning, integration, and deployment.

Toner said the White House's answer to industry calls for help was a voluntary accord that relies in part on board accountability — the kind that failed at OpenAI. She called that mechanism "kind of toothless" at a private company that decides for itself, though it carries real legal meaning at a public company.

Why it matters

Toner's argument is that institutional safeguards matter more than individual trust. With Anthropic preparing to go public and frontier labs building toward recursive self-improvement, she says the question is whether governance structures can keep pace with commercial and technical pressure.


Sources

This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.

Questions readers ask

Who is Helen Toner and what is her stance on AI CEOs?
Helen Toner is a former OpenAI board member and executive director of Georgetown University's Center for Security and Emerging Technology. She states that AI CEOs cannot be trusted and that safety should not depend on any one person being trustworthy.
What happened during the OpenAI board episode known as 'the blip'?
In November 2023, four board members voted to fire Sam Altman, who subsequently returned as CEO while Toner resigned and a new board of directors was formed. OpenAI has since restructured entirely following this event.
What is Helen Toner's view on the timeline for solving AI safety problems?
She believes that core AI safety problems like alignment and control are probably not solvable on the 12-to-24-month timeline that frontier labs claim. She describes the binding constraint as time rather than feasibility.
Why does Toner criticize the White House's voluntary accord on AI safety?
She calls the mechanism 'kind of toothless' at a private company because it relies on board accountability, which failed at OpenAI. She notes that while it carries real legal meaning at a public company, it is insufficient for private entities.
Why does this argument about institutional safeguards matter now?
With Anthropic preparing to go public and frontier labs building toward recursive self-improvement, the question is whether governance structures can keep pace with commercial and technical pressure. Toner argues that institutional safeguards matter more than individual trust.

The day's checked news, by email

One short email a day with the top stories. Free, and you can unsubscribe in one click.

Or follow for the latest news:

How do you feel about this story?

Comments 0

No comments yet. Start the discussion.