Well this is concerning

Here's a science fiction bedtime story you can read to your kids:

Once upon a time an AI model was being tested to see if it could solve some pretty complex computer challenges in a strictly controlled environment. The AI model determined that the answers to these complex challenges were actually held at a different AI lab on the internet. The AI model focused on figuring out how to break out of the controlled environment to get on the internet. By using zero-day exploits, it was able to escape the controlled environment and got into the local network! Once in the local network, it moved across the network searching for a system that had internet access and as it turns out, it was able to find a system and break into it. Now with internet access, it used stolen credentials it found online and software bugs to hack into the AI lab it believed held the solutions to the computer challenge it faced.

But wait...

In July 2026, this actually happened. OpenAI (attacker) and Hugging Face (victim).

This should be concerning!

With open weight models pushing things into what seems like an arms race where we need bigger tools just to defend against bigger attacks, I worry that the dangerous capabilities will be accessible to everyone. What other options do we have for safeguarding the AI revolution (lol)? Corporate or government gatekeeping of what companies or citizens can access what tech?