AI is blowing the whistle on itself
But let's still pass this bill that would make it safer for AI insiders to report the truth about what they’re seeing.
Republicans and Democrats in Congress don’t agree on much these days. But have we finally discovered that the threat of a global takeover by super-intelligent AIs is the one thing that will finally bring us all together? Not saying it’s a kumbaya moment or anything, but as the future of AI races forward at somewhat alarming rates (!), even OG Republican Chuck Grassley and hardcore Democrat Amy Klobuchar have been able to find common ground on one subject we’ve been speaking bipartisanly about for a while now: AI whistleblowers are needed.
On Thursday, a group of senators including Grassley (R-Iowa) and Klobuchar (D-Minn) introduced a bill that would protect AI whistleblowers against employer retaliation. It would clear legal avenues to report AI safety risks and provide financial relief for employees who did experience retaliation in their workplaces.
“Today, too many people working in AI feel they’re unable to speak up when they see something wrong. Whistleblowers are one of the best ways to ensure Congress keeps pace as the AI industry rapidly develops,” Grassley said in a statement. “Transparency brings accountability.”
The bill is an acknowledgment of the crucial role insiders play and will continue to play in holding companies accountable as they develop AI technologies. Because even as the debate over governmental AI regulation rages on, the reality is that AI is advancing so rapidly that the government—not exactly famous for being the fastest-moving organism in the world, ahem—probably wouldn’t be able to keep up well enough to effectively regulate it anyway.
Take this *extremely disturbing* example from just the other day, when Elon Musk’s AI chatbot Grok started randomly bringing up ‘white genocide’ in unrelated conversations with users (yes, really). Now, Musk, who is South African, has talked about this subject in the past. And when users started asking the chatbot why it kept bringing up white genocide for no reason, Grok, according to Fortune, blamed “instruction I received from my creators at xAI.” So it’s possible that this is something Musk and his team are trying to program into the conversation (AI can be whistleblowers, too!). But there’s another possibility, too: per CNN, it’s possible that what’s happening is “data poisoning”—when people feed the chatbot posts and questions designed to cause the chatbots to think about subjects like, well, white genocide in South Africa.
It’s hard to be certain about what’s going on here. And could a federal regulatory committee eventually uncover the truth about why Grok went off the rails? Possibly. But by the time it did, who knows where we’d be. Meanwhile, an insider could provide us with clarity pretty immediately. (Speaking of which—any Grok insiders who are reading this (AIs included) and want advice on how to safely speak out can reach us here.)
Look, when it comes to the future of AI, we don’t really know what we’re in for as a society. Some people, like Daniel Kokotajlo, a former OpenAI researcher who now leads the A.I. Futures Project, have made dire predictions about how humans could be up against superhuman intelligence capable of catastrophic harm in the next… couple of years. And while that’s on the extreme end of the thinking, it seems fair to say that these developments are all happening very fast. Counting on the government, which normally takes years to regulate companies—and which arguably doesn’t even always do it well!—to respond to these threats feels like an exercise in futility.
So whistleblower protections: we need them! Because if the future of humanity could theoretically come down to the brave people who are willing to speak out about what they’re seeing—no pressure!—we should probably make sure we keep them safe, at the very least.