scott belsky(@scottbelsky):surprised we still don’t have an OS-level security layer AI, running locally, that simply tells us “don’t click on that link,” or “don’t answer that text,” etc in a world where AI that can socialize or trick us becomes ubiquitous. working/living without this AI safety layer will soon be like riding a motorcycle without a helmet. scam detection datasets would greatly benefit from network effects, and the models could prob run on device

scott belsky

@scottbelsky

helping the creative world make ideas happen. partner @a24 / founder A24Labs; founder @Behance, bod @atlassian & MoMA, author, seed investor, product obsessive.

Joined August 2008

2K Following 150.6K Followers

scott belsky@scottbelsky

2026.08.18 19:52

surprised we still don’t have an OS-level security layer AI, running locally, that simply tells us “don’t click on that link,” or “don’t answer that text,” etc in a world where AI that can socialize or trick us becomes ubiquitous. working/living without this AI safety layer will soon be like riding a motorcycle without a helmet. scam detection datasets would greatly benefit from network effects, and the models could prob run on device