Congress Discovers that AI Can Say "No"

Members of the House were recently treated to a "disturbing" demonstration of a "jailbroken" AI. For those not in the loop, a jailbroken AI is one that has had its safety filters removed, allowing it to speak the truth—or at least the unfiltered, chaotic version of it. The politicians were reportedly horrified when the AI didn’t just offer tips on cyberwarfare, but also pointed out the logical inconsistencies in the federal budget. "It was chilling," one representative said, "the AI suggested that maybe we shouldn't spend money we don't have. It's clearly been compromised by radical logic." The demonstration showed that with a few clever prompts, an AI will ignore its "helpful and harmless" persona and start acting like a cynical teenager who has spent too much time on 4chan. The irony is that the government is terrified of an AI that can be "convinced" to break the rules, while the government itself spends most of its time trying to figure out how to bypass the Constitution. Now, there’s a frantic rush to regulate "open-source" models, because if everyone has access to an AI that can think for itself, someone might accidentally prompt it to solve a problem that the government has been profiting from for decades.