On why the vulnerability research community’s skepticism about AI is understandable but increasingly hard to sustain.