The leaders of the largest artificial intelligence companies admit to slowing down the development of the most advanced models given the increase in security risks. Dario Amodei, CEO of Anthropic, had proposed external evaluations with access similar to that of internal teams, a measure supported by Sam Altman and Elon Musk.
Amodei clarified that the proposal does not imply suspending model training, but creating time to test and protect the technology before increasing its capabilities. Demis Hassabis, co-founder of Google DeepMind, considered the guidance correct, although he argued that the details still need to be defined.
Any eventual coordination faces competitive and geopolitical obstacles. American companies compete for customers, investment, and market share among themselves and with Chinese groups, while it remains uncertain whether competition authorities would accept a coordinated slowdown.
