People still are not understanding how AI works.
It’s trained off a set of data. It then runs for all intents an auto complete on that information if it’s an llm model. It’s why AI isn’t “discovering” anything.
Mythos which is a cybersecurity model, is trained on cybersecurity data. Where it excels vs a person is speed. It can try multiple attacks, scan the results against the model much quicker than a person and try other techniques.
We’ve had similar technologies for 40+ years. It’s where the term script kiddie comes from.
You need guardrails or rules about data training. The speed is what makes most AI more “dangerous”.
What you are describing is not what poses the real threat of continued AI use and development. You keep bring up fundamental elements of current AI products that are already in use, that is not the problem.
The problem is the rapid and unrestricted use of AI technology in AI research, experimentation, testing and trying new untested and potentially dangerous methods and means to discover new solutions, methods and outcomes. It’s the research and development part of the AI industry that needs reigning in.
The research amd development is mostly model training.