Mythos / Fable: I decided not to share the posts in question so this would not look like targeted bashing (it is not) of accounts I follow and whose content I generally appreciate.
But many of them have a real problem with their reasoning: they turn the reality of cybersecurity completely upside down.
Vulnerabilities do not appear because a model finds them. They already exist. Private stockpiles of flaws have existed for decades among researchers, offensive security companies, states and criminal groups.
The difference today is that AI can speed up their discovery and exploitation, but that acceleration applies not just to attackers, but to defences too!
Banning or limiting the best models means depriving defenders of mitigation capabilities while offensive actors continue using their own tools, their own models, their own exploitation chains and their own stockpiles of vulnerabilities.
A model, even one with "advanced" cyber capabilities, *is not* a nuclear weapon.
It lets us find and fix flaws faster, flaws that would otherwise remain quietly exploitable for years.
Security will not come from artificially imposed ignorance among defenders, but from systems that are better designed, better tested and better patched, and tools powerful enough to finally do the job at scale.
All my links, projects and social media: https://all.bo