OpenAI releases June 2025 Threat Report

The report is interesting in that shows all the typical ways humans are awful.

But none of their efforts (which they call "disruptions") have anything to do with AI or models. They're just trying to catch bad actors after the fact by analyzing the prompts and replies their tool gave people.

There is no such thing as AI safety, any more than we could have typewriter safety. Bad typewriter content (ie: Ted Kaszinsky) must be assessed by the people's actions and constrained the same way -- not by tool limitations, which isn't really possible.

OpenAI’s report shows how they are able to detect, disrupt, and expose abusive activity including social engineering, cyber espionage, deceptive employment schemes, covert influence operations and scams. With some interesting examples.