Introducing a neural-symbolic security architecture for bounding jailbreaks and unauthorized capability release, applicable both to open and centralized frontier AI systems: bengoertzel.substack.com/p/un-jailbreak…

@bengoertzel @AnthropicAI

@SingularityNET @bengoertzel @fluence_project adds another layer. Decentralized compute means no point of failure, no central server to hack, and no one company controlling what runs and what doesn’t.

@SingularityNET @bengoertzel the reason most people underestimate jailbreak risks is they're thinking in terms of current tech limitations, not the potential for exponen

@SingularityNET @bengoertzel Neural-symbolic security? Now we’re talking.

@SingularityNET @bengoertzel code or paper?

@SingularityNET @bengoertzel neural symbolic bounds might hold up better

@SingularityNET @bengoertzel great