There is a lot of online debate about the extraordinary cyber capabilities of AI, and how we all ought to manage introducing them to the world. I believe all of these statements are true:

1. Frontier AI models are exponentially more capable of hacking into things (external organisations, digital infrastructure) and out of things (sandboxes) than humans ever have been.
2. AI models with open weights (downloadable and usable by anyone) are catching up to frontier capabilities at a very quick rate.
3. In cyber, attacking has always been easier than defending. AI is now widening that gap exponentially.
4. These trends cause real concerns for national level security threats (and any other level).

But,

• AI model guardrails (restrictions) around cyber will slow down defenders, while not slowing down attackers (see point 3).
• Phrases like "Globally slowing down the development of AI" are only ever heard in the PR departments of AI companies, the labs can't because it would be suicide.
• Regardless of cyber, US frontier labs would have extreme financial benefit from any type of open weight restrictions. Or restrictions around using models from other countries.
• This benefit heavily influences what they say.
• They are coming out with many stories about how their models have insane hacking capabilities, and they can't even keep track themselves. The models are very scary and we all need to be careful (and protected from evil).

Prometheus

Prometheus stole fire from the gods and gave it to humanity, enabling civilisation, technology, and progress. The myth represents both the power of knowledge and the price of defying authority.

Let's imagine the world of Prometheus: OpenAI warns autonomous hacks are ‘watershed moment for computer security’ Openaius warns an uncontrolled fire can burn down whole villages all at once, a watershed moment for global safety.

When fire gets newly introduced into the world, we're all going to need to adapt fast. Fires destroy and kill, and they can grow quickly and spin out of control. Even to this day fire is very dangerous, as I'm writing this (by hand) big parts of the US and Europe are on fire with massive forest fires. Wouldn't we be better off without fire completely?

Project Glasswing

Zeus saw Prometheus giving fire to humans as an act of theft and direct defiance of divine authority. He feared that fire would make humanity too capable and independent, so he punished Prometheus by chaining him to a rock, where an eagle ate his regenerating liver each day. He also punished humanity by sending Pandora and her jar of suffering.

Project Glasswing is an urgent defensive cybersecurity initiative launched by Anthropic in April 2026, built around their unreleased AI model Claude Mythos Preview. It provides vetted partners including major tech and cloud providers with controlled access to find and fix critical software flaws before bad actors can exploit them.

Should mere mortals (such as non-vetted partners or non-Americans) be denied access to fire? For they might burn down the villages.