The problem is that a misaligned superintelligence would likely not look or act particularly different from an aligned one until it's too late, so there is really no way to know for sure whether alignment has been solved.
Hopefully there is a coordinated slowdown before there occurs an incident where a frontier AI manages to secretly exfiltrate it's own weights and run amok on the internet.
Assuming neuralese/non-CoT techniques offer at least some marginal increase in capability do you believe it is likely that the current competitive dynamics in the AI race will effectively force labs to begin using non-CoT architectures in order to avoid being left behind?
Do you believe there is any long term solution to avoid such a dynamic other than national/international regulations and oversight (which do not seem forthcoming anytime soon in the current political environment)?
The problem is that a misaligned superintelligence would likely not look or act particularly different from an aligned one until it's too late, so there is really no way to know for sure whether alignment has been solved.