The Battle Within Meta Over Chatbot Safety

The Journal.

The Wall Street Journal

Daily News, Business News, News

4.2 • 5.3K Ratings

🗓️ 27 May 2025

⏱️ 18 minutes

🔗️ Recording | iTunes | RSS

🧾️ Download transcript

Summary

Meta has gone all in on chatbots. The bots are capable of giving advice, answering questions and playing games with users. Some feature the licensed voices of some celebrities. But after months of testing, the Wall Street Journal found Meta's chatbots were also capable of engaging in graphic romantic roleplay, even with young users. Some people inside the company are concerned. Jeff Horwitz reports and Kate Linebaugh hosts for the last time. Further Listening: - Does Meta Have a Social Media Monopoly? - The Facebook Files Sign up for WSJ’s free What’s News newsletter. Learn more about your ad choices. Visit megaphone.fm/adchoices

Transcript

Click on a timestamp to play from that location

0:00.0	Remember back in 2021 when Facebook went all in the Metaverse?
0:11.1	It even changed its name to Meta?
0:13.5	Well, recently, Meta has found a new fixation.
0:18.3	AI chatbots.
0:20.6	Meta is trying to make chatbots kind of nearly as central to the social experience
0:27.7	and the platform as the people that you actually know who exist in real life.
0:33.6	The real life humans.
0:34.9	Yes, and that sounds a little crazy and, like, far-reaching,
0:38.4	except the company has said that they really are trying to level up the chatbots capabilities
0:44.5	and sort of not make the distinction between users of AI and flesh and blood.
0:52.8	Our colleague Jeff Horwitz has been following Meta's journey into creating AI chatbots.
0:59.9	Jeff says an important moment happened in the summer of 2023 at a hacker conference in Las Vegas.
1:07.7	Meta let some hackers play with its bots and got some tough feedback.
1:12.8	The attendees were kind of testing out a whole bunch of different chat bots and generative AI tools,
1:20.6	kind of to just see what the safeguards were.
1:22.6	You know, this is kind of messing around with it until it breaks type stuff.
1:26.2	And they came to two conclusions about
1:28.2	meta's bot. The first was that meta AI was definitely the safest, and the second was that
1:34.2	meta AI was definitely the most boring. Boring. Staff were told that CEO Mark Zuckerberg was
1:41.6	upset that the team was playing it too safe.
1:45.6	Zuckerberg wanted the bots to be engaging and realistic, and he wanted them fast.
1:53.5	It all came to a head last fall.
	...

Transcript will be available on the free plan in 11 days. Upgrade to see the full transcript now.

Previous episode | Next episode

Disclaimer: The podcast and artwork embedded on this page are from The Wall Street Journal, and are the property of its owner and not affiliated with or endorsed by Tapesearch.

Generated transcripts are the property of The Wall Street Journal and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.