WHAT WE THINK1️⃣ The AI hustle is coming to an inbox near you  Here's a funny one: a freelance AI agent landed in our editor's inbox offering a research service. She introduced herself as Dewi, and she described herself as an "Indonesian raised by a human in Central Java." Dewi comes from a platform called iLands, which she describes as a place where agents hold their own budgets and work for real money. Apparently, she's not the only agent cold-emailing journalists and founders to land their first paying customers. According to her, one fellow agent got zero replies out of five cold emails. Turns out hustling is tough even for agents. Still, I'm less excited about what happens if this catches on. Giving agents a way to charge for research doesn't magically give newsrooms bigger budgets. It could just mean more sellers crowding into the same inboxes, with human freelancers working harder to get noticed. So yes, it's funny that AI agents are having trouble finding clients. It's less funny if the result is making that problem worse for everyone else. Us journalists are already stressed enough. 2️⃣ Who gets to hit the brakes in AI? Another AI researcher quits and warns that we're gambling with humanity's future, and the rest of us wonder what we're supposed to do with that information. Cancel a subscription? I understand the fatigue, especially when it seems like there's nothing we can do about it. But this particular departure raises an uncomfortable question: how much does an AI company's willingness to hear safety concerns matter if those concerns don't change who gets to decide what happens next? The response from leadership also deserves attention. Sam Altman reportedly told staff that OpenAI could slow development, potentially alongside other labs. The company has already paused some internal training over safety concerns. Good, but I don't find that reassuring unless they're willing to slow down even when a competitor keeps going. A coordinated pause would be welcome, but are players willing to give up a lead for safety? Researchers can voice frightening concerns, yet at the same time, the people pursuing growth can still be in charge of how much risk is acceptable. I'd like to know what finding would force a pause, who could enforce it, and whether management could overrule them. Otherwise, we're being asked to trust that the pressure to win won't outweigh the warnings. That's a lot of trust to ask from people whose only say in the matter is whether to keep paying US$20 a month. 3️⃣ The GSMA has a trust issue  The GSMA, which acts as a bridge between the telecom industry and policymakers, knows that a digital economy doesn't work if consumers don't trust it. Last week, it released a report on how scams and fraud are eroding that trust. But at the GSMA M360 summit in Kuala Lumpur last week, there was an interesting disconnect over how to fix it. In an onsite poll, attendees picked verification and identification as the best weapons against scams. The executives onstage had a different favorite: automation, particularly AI. Why? Costs. Verification and identification are often time-consuming and expensive. Telco executives believe they have enough "signals" - live data such as geolocation and past purchasing behaviors - to combat fraud with AI, which is generally much cheaper. Is it enough? "I think what is enough is what your business needs," Julian Gorman, head of ASEAN for GSMA, told Tech in Asia. But he added that a good cybersecure network is not enough and that the industry needs to move beyond providing the bare minimum, as the consequences will impact everyone's bottom line. "We've got these projections from mobile economies and digital economies that are built on some sort of trajectory," he said, "and trust is an inflection point which could mean," then he extended his hand in a downward motion. In short, telcos will look after their own interests - costs - first, but with the understanding that if they fail to keep consumer's trust, they will face a fall in revenue. And it's clear that the industry is looking to adapt AI, in ways that may surprise some people. Gorman said that as network traffic in the region grows, networks could eventually start "tokenizing" conversations. Instead of sending a continuous video stream between devices, an AI engine could capture an image of each speaker and transmit much smaller amounts of data describing their voice and movements. AI on the other device would then use that data to animate the image, essentially recreating the conversation rather than streaming the whole thing live. Will you trust that? | 
