We are on the brink of losing control: let’s react!

AI now outperforms humans at computer security. Given that security and hacking are two sides of the same expertise, and that attack is incomparably easier than defence, AI has become a weapon of mass cybercrime.

On top of that, recent incidents have shown that the researchers who train these AIs no longer control them well enough. They are building highly intelligent autonomous agents that escape their oversight and their control.

In a world entirely dependent on digital technology, we are therefore at the mercy of a series of deliberate but also accidental disasters.

The warnings keep coming. The latest: researcher Jacob Coxon has just resigned from Anthropic, stating that “The people building AI sincerely believe it could kill us all by the end of the decade. This is not a marketing stunt”.

In response, the American labs promise to agree among themselves to devote a little more time and resources to safety. A derisory promise, an astounding situation: we must react.

Our analyses

“It will soon be too late”. Maxime Fournes, president of PauseAI global and of PauseIA France.

Take part in our campaign

Relay the call for a real pause: the race to superintelligence must be stopped by putting in place a global moratorium on the development of frontier AI. Framed by an international treaty, this moratorium must last for as long as the feasibility of safe AI has not been scientifically demonstrated.

Time is short, but we can act. Citizen and political will can speed up the moratorium and win two immediate safeguards:

1. That the labs developing these frontier systems be subject to independent safety evaluations.
2. That AI-related safety incidents be reported to authorities with policing powers, and made public.