It’s good that these security bugs get found by AI. Linux and researchers make heavy use of AI to detect these, also for new code. The next years will be hard, with many findings, until AI has maxed out all the patterns it can learn. In the long run it will lead to less bugs. We already had a similar situation when fuzzers became a thing.
You can definitely restrict access to sockets via LSMs. Whether these exploits will work depends on the security context an attacker is in.