Unfortunately there is a risk. What you read mostly is about LLMs, but the rapid evolution in hardware and research also benefits other neural networks applications. The government and military are as worried as when the atomic weapons begun to be developed: they don't want to fall behind so the global military scales won't be tipped against you. Along with everything else, China is autocratic which makes them more agile in big changes: see if there's any protests about data centers there for example. Drones as an example: China is making them very cheap and to counter them is currently expensive. If they manage to make these drones fully autonomous too, it's a big threat. That also applies to other weapons operation and development. The idea is to one day not need a human to plan and execute a military operation. And notice that using AI for this isn't comparable to nuclear weapons in the sense you potentially will screw up the whole planet if you use it.
Despite despising this A"I" circus, some arguments in the sea of discussions are valid, IMHO.