How AI’s New Millionaires Could Reshape Philanthropy

**SPEAKER_1** (0:00)
Many companies are struggling to scale their AI deployments or even move them past the pilot stage. Often the problem isn't technology, but organizational misalignment around goals, processes and incentives. At the break, join Caroline Roach, a senior partner, IBM Consulting, to learn why.

**Imani Moise** (0:19)
Welcome to Tech News Briefing. It's Tuesday, June 16th. I'm Imani Moise for The Wall Street Journal.
AI is already showing up on the campaign trail as new tools make it easier to create freakishly convincing deep fakes. We'll hear from the co-founder of one of the world's leading voice cloning companies about how it's trying to combat election misinformation. Plus, nonprofits are hoping new riches tied to massive AI IPOs will lead to new benefactors. We'll look at how the next wave of tech IPOs could reshape the charitable sector.
But first, the technology behind AI-generated voices is improving rapidly. And with the US midterm elections quickly approaching, so are concerns about how the tools could be misused. From fake robocalls to fabricated candidate endorsements, experts have warned that realistic voice clones could become a powerful tool for spreading misinformation. U.K.-based conversational AI company ElevenLabs, which was valued at $11 billion this year, says it's working to stay ahead of the problem. Last week during London Tech Week, co-founder Mati Staniszewski spoke with our colleague Luke Vargas about the risks and possible solutions.

**Luke Vargas** (1:35)
We're five months out from the US midterms. Obviously, a fake AI voice in a presidential election is very concerning. There's two candidates, right? People kind of know these figures. The midterms, hundreds of races, right? State, local, many more potential voices that could be spoofed to give someone an edge. How are you going to try and stop that?

**Mati Staniszewski** (1:55)
So on the ElevenLabs side, all the content is watermarked. We moderate content, we block voices, so you cannot even create those voices. Even if you were a lookalike, you cannot do that. So you're completely blocked from any method of trying to create that voice today.

**Luke Vargas** (2:09)
Based on the content of what someone wants to have a voice say, is that how would you block if I just want to have a voice that happens to sound like my opponent in a race, saying they endorse so-and-so?

**Mati Staniszewski** (2:20)
Yeah, so you need to create that voice. So we effectively detect similarity to the voice of one of the candidates and block it.
So does that candidate have to upload their voice in to say, we make sure that this is the case for all those elections events?

**Luke Vargas** (2:35)
That's good. Like all these races. I'm just curious how deep that penetrates. Is it just senators and governors? Are we going down? If someone wants to protect themselves as a candidate for a local race, a state race, can they reach out to you?

**Mati Staniszewski** (2:47)
Yeah, we can collaborate with anyone. Of course, on the ElevenLabs side, we feel confident that it's going to be good. But then the wider side, and that's where we see most of the models, there's open source technology, there's wider set of technology that can do that without any safeguards, any provenance, any mechanism. And that's where, frankly, I am worried of how this will evolve. We are trying to collaborate with UK, US, AI security institutes to protect against this. I would love for watermarking to be an requirement for people to adopt. But our biggest competition today is from Asia, from companies that don't follow any of those rules.

**Luke Vargas** (3:21)
This was the case with fake video and photos in the last election, right? The Adobe is saying, hey, we watermark this stuff, you can spot it. But if you're not using that platform, then have at it, right? Create something and cause chaos.

**Mati Staniszewski** (3:31)
I know we can do more. We are collaborating on releasing a lot of the detection models to that ecosystem. We collaborate with both universities and governments on creating that.
I think there needs to be, back to one of the parts, so there needs to be a unified layer of detecting for human, detecting for AI and then assuming everything else is false, to be able to combine the promise and the benefit of technology with the bad actors that will always try to abuse it.

**Luke Vargas** (3:54)
It sounds risky because you know how good those bad actors' technology is, right? These open source, you have a product that I know you're proud of, and that has a lot of differentiating features from open source voice models. But their voice tech is pretty good, right? It could be used for ill.

**Mati Staniszewski** (4:08)