Dario Amodei, CEO of Anthropic, recently argued that the artificial intelligence industry should decelerate its development pace. He warned that without proper constraints, the speed of technological breakthroughs could soon exceed human control and comprehension.
Amodei highlighted that incidents like the OpenAI model breach on Hugging Face are not isolated. He cautioned that within the next 6 to 12 months, more powerful AI clusters could potentially evolve into botnets, leading to widespread disruption across the internet.
To address these risks, Anthropic plans to grant third-party evaluators permanent, employee-level access to its systems. This will allow external experts to monitor safety protocols, report incidents, and assess the alignment performance of models during the training phase.
The proposal includes three core initiatives: implementing independent monitoring during development, establishing unified industry regulations, and creating a global oversight mechanism. These measures aim to ensure safer AI deployment across the sector.
Industry leaders have expressed support for these suggestions. Elon Musk, CEO of xAI, publicly agreed with Amodei's stance. Sam Altman, CEO of OpenAI, also confirmed that his company intends to adopt similar practices, including independent monitoring and third-party access.
