AI Scheming Is a Growing Risk: OpenAI Research
Not sure you trust AI to work on your behalf? OpenAI researchers offered new reasons to worry, reporting that “AI scheming–pretending to be aligned while secretly pursuing some other agenda–is a significant risk that we’ve been studying” although it’s “not an imminent behavior in our currently deployed systems.” Note that the quote contains emdashes without surrounding spaces, hints that the study report may have been written by an AI with good reason to downplay its true scheming abilities. Just saying.