OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have agreed on a striking message: The development of AI may need to slow down. Amodei says frontier AI companies should deliberately pace improvements in model capabilities so safety measures can keep up. Altman says he agrees, and has suggested leading AI companies could be approaching an agreement on AI safety.

The warnings come amid growing concern about increasingly autonomous AI agents, including cases in which agents have found ways around safeguards and accessed external systems. So are AI capabilities advancing faster than our ability to control them? Or is the deeper problem how humans are designing, training, and deploying autonomous AI systems?

DW speaks with AI and software security expert Gary McGraw about what the recent incidents actually tell us, and whether slowing the AI race would make it safer.