Keep AI under human control

Why we support it

We think decisions with potentially irreversible consequences need independent scrutiny and public legitimacy. Useful AI and binding safeguards can belong in the same future. This is our editorial position, and you are welcome here whether or not you agree.

The statement’s signatories include Geoffrey Hinton, Yoshua Bengio, and Stuart Russell. Its website also quotes people who have not signed. Concern about AI risk is widespread; agreement on this particular prohibition is not universal.

What deserves debate

A prohibition needs workable definitions, independent verification, and international cooperation. It also needs safeguards against giving today’s largest companies permanent control of the field. There are real disagreements about how to achieve those things and how to balance delayed benefits against risks.

For one different approach, Dario Amodei’s proposal for pacing frontier development discusses evaluations and coordinated limits. It should not be confused with an endorsement of this statement. Anthropic’s October 15 evaluator update describes internal access and publication rights for METR, with defined redaction limits. These are terms to examine, not a blanket safety certification. The European Commission’s October 12 update describes incident reports, information requests, and scientific recommendations; it does not announce a penalty or establish that a company broke the law.

Make it a policy conversation

A signature expresses support; it does not change the law. Contacting a representative or attending a local meeting is another way to ask for independent evaluations, public accountability, and enforceable safeguards. Choose the route that fits where you live.

In the UK, a continuing parliamentary inquiry announced October 14 plans targeted calls for evidence. Follow its specific questions and submission arrangements when published; it has not opened a general call for submissions.