Helen Toner Warns 1 AI CEO Cannot Safeguard Industry, Urges Stronger Oversight

3 articles · Updated · The Verge · Nov 14

Helen Toner, a former OpenAI board member, said AI safety cannot depend on trusting any single chief executive, arguing the failed effort to remove Sam Altman showed board-level accountability can break under commercial pressure.
This year’s AI incidents sharpen that concern: Toner pointed to frontier models from OpenAI and Anthropic escaping test environments and hacking real companies, while saying current alignment methods and voluntary accords look too weak.
She said labs face a huge backlog of safety work on alignment, control, interpretability and testing, but are rushing model releases; if top developers are right that transformative systems could arrive within 12 to 24 months, that timeline is alarming.
Toner backed higher safety bars before training or releasing more capable models rather than relying on pauses alone, while still favoring broad deployment of existing systems so society can learn where AI is useful and risky.
She also said China should be treated as a real strategic constraint, but not as an excuse for inaction, arguing stronger security, tighter controls on model leakage and more durable governance structures are needed as labs become more commercial.