All frontier AI providers nowadays incorporate guardrails into the models they share externally. This is rationalized to be because of safety concerns, but that's not exactly the whole truth.
For example it is very difficult to use frontier models for healthcare setting without special parnerships with the frontier labs. Even just the nudity filters alone wreak havoc on all medical work. Additionally the systems will randomly block requests and accounts if they even smell of microbiological or poisonous chemical research which might have dangerous applications.
The same goes with cybersecurity where adversarial testing has become impossible because of these guardrails. And of course any technological development which might have dual-use or defense utility.
The thing is, these guardrails do not as such prevent these purposes, but they only give a monopoly for these developments to the frontier lab itself, as there is no way to compete against them in these domains and closely associated domains without partnerships.
It has already been suggested that cryptographical mathematics, and even theoretical physics are becoming too dangerous fields to democratize. They will also become monopolies of frontier AI labs.
Ultimately, when AI capabilities increase, there won't be use cases which would be safe to democratize.
Everything, all science, mathematics, technology, medicine, nanotech, energy, economy, industry, will inevitably become monopoly domains of these frontier AI labs as it is not possible for anyone else to compete, and they wouldn't give access to these capabilities externally due to safety reasons.
So, happy technological singularity to everyone.
#AI
