I think npm can use chatgpt/claude for each published package to detect these types of attack? And if it sees they can restrict the package from making any changes.