But this isn't fundamentally different than people writing scripts that attack such endpoints (or that attempt to use them benignly, but fail). This is still a

But this isn't fundamentally different than people writing scripts that attack such endpoints (or that attempt to use them benignly, but fail). This is still a "human malice and error" problem, not a "dangerous AI" problem.