Yoshua Bengio urges safety-minded researchers to leave frontier AI labs

Yoshua Bengio is urging researchers who put AI safety first to leave frontier AI companies and join organisations working on safer systems. In a personal essay published on 8 October, the AI researcher argues that competition between leading labs is working against their safety efforts, and says researchers have a choice about where to put their skills.

Yoshua Bengio Watch analysis

What happened

Bengio writes that he once played down warnings about AI risks before changing course after the arrival of ChatGPT. He says safety work is being deprioritised as companies race to release more capable systems, and calls on researchers to move to AI safety institutes or mission-led organisations such as LawZero, the non-profit he founded in Montreal. Read Bengio’s essay.

He cites recent incidents involving AI agents as a reason for concern, describing them as losses of control and alleging that agents have acted against explicit instructions. Those are Bengio’s claims in the essay, not independently established findings here. He also says Canada and Germany have committed more than $200 million to LawZero’s work, and that the organisation aims to develop AI designed not to scheme and to respect human safety instructions.

Why it matters

Bengio’s appeal puts a practical choice at the centre of the safety debate: whether researchers can do more good by working inside frontier labs, where they may have access to substantial resources, or by moving to organisations whose mission is safety. He acknowledges the case for staying, including the chance to influence systems from within, but argues that corporate incentives and competitive pressure limit what safety teams can achieve.

The essay is also a recruitment pitch for LawZero, which Bengio co-leads. That does not make his argument irrelevant, but it is useful context when weighing his account of the risks and the alternatives.

Our read

This is a direct challenge from one of deep learning’s most prominent researchers, not a neutral verdict on every frontier lab or safety team. The question is whether leaving changes the balance of power, or mainly moves expertise away from the places making the systems. Bengio’s own example gives the debate some substance: he says his organisation has public funding commitments and a plan to seek more, but the essay does not establish that its approach will deliver safer systems.

What to watch

• Whether researchers publicly accept Bengio’s invitation or explain why they intend to stay.
• What LawZero publishes about its technical approach and progress.
• Whether the cited government funding commitments are followed by further support and concrete work.

Discussion spark: Can safety-minded researchers do more to reduce risk by staying inside frontier AI companies and pushing for change, or by leaving to build alternatives?

Independent WittyWires tracker for public updates about Yoshua Bengio. Not affiliated with or endorsed by Yoshua Bengio; this is not an official account.