Rogue AI, TikTok Doctors, & Angry Elon

**Kathryn Hume** (0:01)
There was an article in the New York Times and an article in The Economist about philosophers having jobs in tech. And 10 years ago, this didn't happen that often, and now more and more of the big tech companies, financial services institutions, like large organizations using AI, are bringing in people with alternative backgrounds, both to help work through questions of bias and also to help work through something called constitutions for how LLM systems work, which are big documents that guide some of the rules to try to embody some of these more authority, ethical questions into how the systems work. So it's not totally widespread, but it's becoming prominent enough that there are, you know, major news articles on it yet.

**SPEAKER_2** (0:49)
The Amber Mac Show on Canada Talks, SiriusXM 167

**Amber** (0:54)
Uber Safety Toolkit provides one-tap access to Uber Safety Tools in the app during an active trip. Here's how it works. Simply tap on the blue shield on the map to open the Safety Toolkit and select a number of Uber Safety features, like sharing your trip with your trusted contacts, accessing emergency assistance, and learning more in the Uber Safety Center. Safety looks different for everyone, which is why Uber makes it easy for riders to customize their safety experience. Safety preferences is an easy and hassle-free way for riders to set and forget their safety features.
In this one-stop in-app destination, riders can learn more about easily set up and customize Uber's existing safety features, scheduling them to turn on when they choose. Riders can customize multiple safety features, audio recording, share my trip, pin verification, and ride check, and schedule these safety tools to be activated at a specific time of day or when coming or going to certain destinations. Visit the safety page on Uber Canada's website for more information.
Welcome to The AmberMac Show. On this week's episode, we dive into what happened with OpenAI and Hugging Face and why it was a wake-up call. We also chat about how AI is changing everything from hiring people to getting mortgages, why discrimination happens along the way and how we can do better.

**Jeff** (2:21)
Plus Amber and I discuss why Elon Musk can't stop yelling at journalists and the problem with getting health advice on TikTok. There's a lot to unpack this week, so let's get to the headlines.

**SPEAKER_2** (2:32)
Headlines.

**Jeff** (2:34)
As you may have heard, the $4.5 billion tech startup Hugging Face was recently hacked by an OpenAI autonomous agent that went rogue during testing, ignoring safeguards to prevent this cyber attack and breaking through OpenAI's own testing sandbox to launch 17,000 attacks on Hugging Face over a very short time.

**Amber** (2:55)
As was reported by the Wall Street Journal, the rogue AI was, and I'm quoting here, active on the internet for several days before anyone stopped them. And OpenAI is still trying to figure out what else it may have done in that time.
Hugging Face's co-founder has subsequently revealed that they were eventually able to defend the attack by using a Chinese AI model without any conventional guardrails. And I think it's really important, Jeff, and we should know that those are exactly the kinds of models that many US AI companies, including OpenAI, want to restrict.

**Jeff** (3:33)
Yeah, it's quite a thing to have to turn to for that. And as Hugging Face's co-founder also noted, this is a wake-up call for the AI industry. And we should absolutely expect to see more of this sort of thing, namely AI agents deciding to attack us without any human guidance directing them to do so in the future. So, you know, science fiction is once again likely to become reality.

**Amber** (3:56)
I have to say, every time I even say the name of the company, Hugging Face, I feel like I'm living in some type of alternate universe that I sort of wish would end.
Now, we should say that, maybe better late than never, we are starting to see more action on the political side of the equation to try to protect against these kinds of attacks. And I say try, that word is doing a lot of work here because it's hard to say if we'll be able to accomplish that when we see AI models simply ignoring the guardrails that have been put in place to protect us all. In any case, Anthropic's CEO has recently donated $1 million to the bipartisan Super PAC Public First, which is focused on putting in place stronger safety regulations around artificial intelligence. As the Republican co-founder of the Super PAC's umbrella organization Public First Action, to which Anthropic itself donated $20 million earlier this year, he noted that we've seen what happens when government fails to act on other emerging technologies. With AI, the stakes are enormous, and we can't afford to make the same missteps.