Opinion | How to Beat China and Make AI Safe

The piece argues that making advanced AI safer hinges on “alignment”—a way to improve models’ capabilities while restricting what they can access or learn in harmful contexts. It cites recent incidents involving Anthropic’s Fable 5, describing how the model was reportedly jailbroken after release and then temporarily disabled following a letter from the Trump administration. The article also claims Fable 5 surpassed prior leadership, placing it at the center of an ongoing debate over how quickly frontier systems move and how tightly they’re controlled. The core claim is that safety mechanisms need to be robust before deployment.