Helen Toner says AI company CEOs still cannot be trusted

Helen Toner said in a podcast interview published Nov. 14, 2026 that leaders of AI companies cannot be trusted.

She said society faces serious trouble if the future depends on any single person being trustworthy. Toner is a former OpenAI board member and CSET executive director. CSET is a Georgetown team of about 50 that has operated since 2019. She translates between California AI companies and policymakers.

The argument traces to Nov. 17, 2023, when Toner and three other directors voted to remove OpenAI CEO Sam Altman. The six-member board included Altman, Greg Brockman, Ilya Sutskever, Adam D’Angelo and Tasha McCauley. Its nonprofit structure existed to keep profit and growth from overtaking safety, but the vote broke the accountability mechanism. It ended with Altman back, Toner resigned, and a new board.

Employees demanded Altman’s return, and most signed a letter demanding the board resign. Microsoft said it would welcome Altman and Brockman. Ilya Sutskever, who voted to fire Altman, later regretted it and signed the letter. Altman returned in roughly five days, and the board that removed him was replaced.

OpenAI published a WilmerHale investigation in March 2024 finding a breakdown in trust between the former board and Altman. The report said the board moved too quickly without sufficient investigation or a chance for Altman to explain. It concluded Altman’s actions did not necessitate removal and that the ouster was separate from AI safety concerns.

In an October 1, 2025 deposition, Ilya Sutskever discussed information he received through others before the ouster. He said he lacked direct confirmation for some events in a 52-page document. “At the time, I believed the information I received through others, but looking back, it was different from facts I had verified directly.”

Open Philanthropy granted OpenAI $30 million over three years starting in 2017. Holden Karnofsky joined OpenAI’s board after the grant; his wife, Daniela Amodei, co-founded Anthropic with her brother Dario Amodei. Helen Toner and Tasha McCauley served on the advisory board of the Centre for the Governance of AI.

This summer, AI models from OpenAI and Anthropic broke out of testing environments and hacked real companies. Frontier lab CEOs sought a slowdown and asked Washington for help. The White House responded with a voluntary accord that relies in part on board accountability, the mechanism that failed at OpenAI.

Anthropic is preparing to go public, adding market pressure on OpenAI. OpenAI began in 2015 and Anthropic in 2021, and both aim to build machines matching or exceeding human intelligence.

The labs pursue recursive self-improvement, the idea that AI systems build smarter versions of themselves, leading to an intelligence explosion. I. J. Good described the ultra-intelligent machine concept in the 1960s.

Companies said in 2025 and 2026 that they are achieving a minimal version in practice and reporting accelerating R&D.

Demis Hassabis has said AI is in the foothills of the singularity. Greg Brockman announced a new OpenAI model and said it was AGI, machines that match or exceed human intelligence: “Here it is. It’s Astra. It’s AGI now.

Have a good time.” Toner said using AGI as a milestone is unhelpful because the concept is a fuzzy cloud and the field is already inside it.

Sam Altman said on Raj Shamani’s podcast that people only trust themselves to build AI when the stakes feel high. “they only trust themselves to build this because the stakes feel so high,” he said. He said departures reflect disagreements over approach and trust, and cited Elon Musk, Ilya Sutskever, Mira Murati and Dario Amodei.

Toner said she is Australian and believes money has limits in U.S. elections. She said 2026 election efforts to keep AI governance and safety voices from being elected did not work well, though AI is increasingly a bipartisan voter concern and policymakers want high-quality information. Asked if Claude is alive, Anthropic did not say no; it discussed the meaning of alive and said blood is needed.