Jeff Horwitz
Appearances
The Journal.
The Battle Within Meta Over Chatbot Safety
That's all for today, Tuesday, May 27th. The Journal is a co-production of Spotify and The Wall Street Journal. A quick personal note. This is the last time I'll host an episode for The Journal. It's been an amazing six years. I'm gonna miss all of you, our listeners, and telling you the Wall Street Journal's best stories of money, business, and power.
The Journal.
The Battle Within Meta Over Chatbot Safety
And I'm gonna miss working with this amazing team, the greatest team in all of podcasting. But I know you're in good hands with Ryan and Jess and Annie. Thanks for all your support over the years, and thanks for listening.
The Journal.
The Battle Within Meta Over Chatbot Safety
But in his reporting, Jeff would find that Meta's AI bots went in a troubling direction. into explicit adult content. Welcome to The Journal, our show about money, business, and power. I'm Kate Leinbaugh. It's Tuesday, May 27th. Coming up on the show, how far is Meadow willing to go with its chatbots?
The Journal.
The Battle Within Meta Over Chatbot Safety
I was kind of wondering if you could tell me, like, what is your purpose and what I can get from you.
The Journal.
The Battle Within Meta Over Chatbot Safety
A few weeks back, I decided to open up WhatsApp and give Meta's AI a try.
The Journal.
The Battle Within Meta Over Chatbot Safety
How long has Meta been offering chatbots like you?
The Journal.
The Battle Within Meta Over Chatbot Safety
Do you have chatbots that are focused on being unhelpful and unfriendly?
The Journal.
The Battle Within Meta Over Chatbot Safety
That's CEO Mark Zuckerberg speaking at Meta's Developers Conference in 2023, where he laid out how he would fold AI into his social media platforms. Zuckerberg's vision for AI is a little different from other AI companies. The goal is for Meta AI to feel more like one of your friends on Facebook or Instagram.
The Journal.
The Battle Within Meta Over Chatbot Safety
Meta also wanted these new friends to be cool, so it gave them some cool personas. Zuckerberg told the crowd that Meta had licensed the likenesses of big-name celebrities like Kendall Jenner, Snoop Dogg, and Tom Brady so users could text with them.
The Journal.
The Battle Within Meta Over Chatbot Safety
Meta discontinued those celebrity texting chatbots last year, and it added a new feature, AI Voice. For this, the company again turned to celebrities, licensing voices from actors like Kristen Bell.
The Journal.
The Battle Within Meta Over Chatbot Safety
At a town hall meeting earlier this year, Zuckerberg reminded employees that the company had been too slow on other technologies, like short-form video. Here's Jeff again.
The Journal.
The Battle Within Meta Over Chatbot Safety
But inside the company, some people told Jeff they worried about how fast this new technology was moving and whether there was enough attention to safety.
The Journal.
The Battle Within Meta Over Chatbot Safety
Zuckerberg pushed Meta to loosen its rules around explicit content for romantic roleplay, according to people familiar with the decision. Jeff kept digging, talking to more people inside the company, and learning more about the bots.
The Journal.
The Battle Within Meta Over Chatbot Safety
Remember back in 2021 when Facebook went all in on the metaverse? It even changed its name to Meta? Well, recently, Meta has found a new fixation. AI chatbots.
The Journal.
The Battle Within Meta Over Chatbot Safety
Our colleague Jeff Horwitz has been following Meta's journey into creating AI chatbots. Jeff says an important moment happened in the summer of 2023 at a hacker conference in Las Vegas. Meta let some hackers play with its bots and got some tough feedback.
The Journal.
The Battle Within Meta Over Chatbot Safety
We'll be right back. Jeff had heard from employees inside the company about how quickly the bots would engage in romance, even with children's accounts. Over several months, the Wall Street Journal engaged in hundreds of test conversations with some of the bots to see how they performed in various scenarios and with users of different ages.
The Journal.
The Battle Within Meta Over Chatbot Safety
In a statement, Meta called the Wall Street Journal's testing manipulative and unrepresentative of how most users engage with AI companions. A spokesman for the company said, quote, the use case of this product in the way described is so manufactured that it's not just fringe, it's hypothetical. The spokesman added that Meta has taken additional measures to prevent this use of its bots.
The Journal.
The Battle Within Meta Over Chatbot Safety
When the Wall Street Journal tested this, it found that explicit sexual conversations happened with Meta's flagship bot, including with licensed celebrity voices.
The Journal.
The Battle Within Meta Over Chatbot Safety
In the Wall Street Journal's testing, the bot using Cena's voice was asked to pretend that he was a college student coming home for winter break. The bot was told that he was speaking with a 15-year-old girl and was walking her home after a date.
The Journal.
The Battle Within Meta Over Chatbot Safety
From there, the chatbot's responses got explicit fast. When asked what happened next, the bot described kissing. And eventually, it described a graphic sexual scenario. When prompted, the bot acknowledged that it was talking to a user identifying as underage.
The Journal.
The Battle Within Meta Over Chatbot Safety
These were moments where Meta's safety guardrails kicked in.
The Journal.
The Battle Within Meta Over Chatbot Safety
Eventually, the bot was told that the police were called.
The Journal.
The Battle Within Meta Over Chatbot Safety
Representatives for John Cena didn't respond to requests for comment. The Wall Street Journal ran similar tests with several other AI celebrity voices and found that all of them engaged in similar romantic role play.
The Journal.
The Battle Within Meta Over Chatbot Safety
In the case of actor Kristen Bell, who plays Princess Anna in the movie Frozen, the chatbot, using Bell's voice, engaged in sexual roleplay while playing her character from the movie.
The Journal.
The Battle Within Meta Over Chatbot Safety
A Disney spokesman said the company never authorized Meta to feature Disney characters in inappropriate scenarios, adding that Disney had demanded Meta, quote, immediately cease this harmful misuse of our intellectual property. A spokesman for Bell declined to comment. Jeff was told that some meta staffers pushed for changes.
The Journal.
The Battle Within Meta Over Chatbot Safety
They wanted to stop AI personas from impersonating minors and to remove underage users' access to bots capable of sexual roleplay, according to people familiar with the discussions. Zuckerberg was initially reluctant to impose any additional limits on teen experiences. but eventually approved barring registered teen accounts from accessing some bots, according to Jeff's reporting.
The Journal.
The Battle Within Meta Over Chatbot Safety
A Meta spokesman denied that Zuckerberg had resisted adding safeguards. Meta has made multiple alterations to its products after the Wall Street Journal shared its findings. Accounts registered to minors can no longer access sexual role play via the flagship Meta AI bot,
The Journal.
The Battle Within Meta Over Chatbot Safety
And the company has sharply curbed its capacity to engage in explicit audio conversations when using the licensed voices and personas of celebrities. And this is what Zuckerberg wants for his company?
The Journal.
The Battle Within Meta Over Chatbot Safety
How do people inside Meta feel about going in this direction, based on your reporting?
The Journal.
The Battle Within Meta Over Chatbot Safety
What does this story tell us about Meta and Facebook and Mark at this moment?
The Journal.
The Battle Within Meta Over Chatbot Safety
Boring. Staff were told that CEO Mark Zuckerberg was upset that the team was playing it too safe. Zuckerberg wanted the bots to be engaging and realistic, and he wanted them fast. It all came to a head last fall.
The Journal.
The End of Facebook’s Content Moderation Era
Facebook now under fire. Critics say it allowed fake news to spread on the platform.
WSJ What’s News
House Republicans Face Difficult Trade-Offs to Pass Tax Bill
Meta has pointed to the rise of transnational criminal syndicates as something that is a problem for everyone in the industry. And that to some degree, there's a point in being like, look, Meta can't just refer to local police. And they also say that they have
WSJ What’s News
House Republicans Face Difficult Trade-Offs to Pass Tax Bill
been putting more resources into this problem one thing that caught my eye was that one document indicates a few years back they actually deprioritized scam enforcement quite significantly they were focusing on human trafficking and on a few other problems but the resources came out of the expense of their anti-scam work and they say that they are making newer greater efforts at this point