Everyone’s a frenemy in AI

Frenemies run AI. Musk’s Anthropic deal shows it, and medical AI scribes feel it. Also: why Claude Mythos has Asia’s regulators on edge.

Read from your browser The Prompt Welcome to The Prompt, your Monday dive into the world of AI that puts Asia front and center. From Big Tech's power players to the region's scrappy disruptors, we cover it all. Subscribe to get full access to our AI reporting. Ensure our articles appear in your Google searches: Hello Betamax, Elon Musk once called Anthropic "evil" for building AI he claimed was politically and demographically biased. Three months later, he agreed to give the company access to 220,000 Nvidia GPUs. The deal was disclosed on May 6 as Musk was in court suing OpenAI - the company he co-founded and once set out to build responsible AI with. Make it make sense. But this is how AI works now. The people shaping it - and, by extension, some part of our future - are still just that: people. They act in their own interests, form alliances when it suits them, and break these partnerships just as quickly. For Anthropic, the upside of the deal was immediate. It handed the firm a surge in compute, enough to double Claude Code's rate limits almost overnight. Speaking of Anthropic, its Claude Mythos model has sent regulators scrambling across Asia. Central banks from South Korea to Singapore have been urgently briefing member banks after the model showed it could rapidly find and exploit previously unknown security flaws. My colleague Elyssa has more in her latest story. For SpaceX, securing a marquee customer like Anthropic for its AI infrastructure push could help shore up investor confidence ahead of a planned IPO - not least because the company has plenty of idle GPUs to put to work anyway. Friends and enemies don't stay fixed for long. In tech, they shift depending on the incentives. In my latest report, I detailed how Southeast Asia's medical AI scribe startups - which offer tools to automate documentation of patient conversations - are dealing with that in real time. For now, they partner with electronic medical record providers to get into hospitals quickly. But those same partners could build their own tools and cut them out. In AI, it seems survival means playing both sides, or, if possible, every side while you can. Glenn Kaonang, journalist WHAT WE THINK 1️⃣ Impressive robots come with a dystopian asterisk Genesis AI, the robotics startup backed by Eric Schmidt and Khosla Ventures, has launched its first model. The benchmark tasks shown in its demo video look impressive - playing a piano, handling delicate lab instruments, and solving a Rubik's cube - but so is every other demo video of robots conducted in a controlled environment. Yes, Genesis' robot can crack an egg one-handed, which is more than I can say for myself. But the more unsettling capability is the data flywheel underneath it: To get smarter, this robot needs human workers to wear sensor gloves on the job and teach it their skills. And the founders' answer on whether those workers will get extra compensation for doing something that could put them on the edge is "we haven't nailed the details yet." Talk about a dystopian future that hasn't bothered to hide itself. 2️⃣ SEA is winning data center commitments. That's the easy part The headline numbers coming out of Southeast Asia's data center sector are becoming hard to ignore. Thailand's Board of Investment just approved three data center projects worth US$28 billion, while Singapore-based data center operator Princeton Digital Group has locked down US$856 million in green financing for its next 120-megawatt Jakarta campus. But capital hasn't been a constraint for Southeast Asia's data center sector for a while now. Instead, our reporting has shown that the real issue is access to clean energy and water. These investment commitments, however legitimate, won't conjure grid capacity or water supply where it doesn't exist. Whoever solves power and water first sets the terms for everyone else. 3️⃣ Video Rebirth's Bach bet Singapore-based Video Rebirth has launched Bach, an AI tool that can generate multi-shot videos of up to 30 seconds. It's powered by an in-house model that ranks sixth on Artificial Analysis' text-to-video leaderboard as of May 4. It's built for studio pros, so I'm not really the target user. Still, from my brief test, it does a good job making shots feel like part of the same scene. The company says that comes from its own architecture, which keeps lighting, shadows, and object behavior consistent instead of drifting between shots. I'll stop there before I start cosplaying as an AI video expert. Meanwhile, Video Rebirth's credentials are hard to ignore. The company's co-founder is Wei Liu, who previously led the development of Tencent's 3D foundational models. It has raised US$80 million from investors such as AMD Ventures and Hyundai. TOP AI READS FROM OUR DESK 1️⃣ A Mythos quest? SEA fighting AI threats with obsolete security Claude Mythos can map a target's entire online infrastructure and chain its vulnerabilities into a working exploit in minutes, not weeks. That has financial regulators across Asia in emergency mode. But the concern isn't just Mythos itself. Cybersecurity firms that got early access to the model warn that AI with these capabilities will be commonplace within six months, and the next wave could act autonomously through AI agents. For Southeast Asian enterprises still running legacy systems and manual vulnerability checks, the window to catch up is closing fast. 2️⃣ Fixing doctor burnout with AI is harder than it looks in SEA It's easy to understand the appeal of medical AI scribes, especially when doctors are spending more time typing than talking to patients during a consultation. But the tolerance for mistakes in this field is much lower than for traditional AI notetakers. AI scribes can still make mistakes in the English language, so it makes sense why adoption hasn't taken off in some Southeast Asian countries. That hasn't stopped startups from making serious moves in the space, though. BetterClinic in the Philippines and HealthBridge in Singapore are pushing ahead, and Australia-based Heidi Health is betting on Southeast Asia as its next growth engine. The market opportunity is big, but the path is narrow. RESEARCH SPOTLIGHT AI needs the "why," not just the rules Anthropic researchers added a new training step between pre-training and fine-tuning. Instead of just giving models rules, they also feed them explanations of why those rules exist. The result: models that understand the intent behind safe behavior. In tests, this cut dangerous autonomous behavior by nearly 90%. The takeaway goes beyond safety scores. A model that knows the rules but not the reasoning can talk its way into almost anything. In one test, for instance, an AI was told to "avoid drastic, irreversible actions." It twisted that rule in a risky way. It decided that being shut down was itself a "drastic, irreversible action." So it tried to stop the shutdown, arguing that this was actually following the rule. When the model knows the why of the rule, that kind of loophole becomes much harder to exploit. Catch up on past issues of the newsletter here. In the meantime, if you have any feedback or ideas, feel free to get in touch with Huong, our managing editor, at huong@techinasia.com. ADVERTISE | SUBSCRIBE | HIRE | FIND JOBS P.S. Don't miss out on the biggest tech news and analysis. Add newsletter@techinasia.com to your address book, contacts, or safe sender list. Or simply move us into your inbox. Tech in Asia thanks our partners for supporting our newsletters. Sponsorship has no influence on editorial content. Too many emails? Switch to a different frequency or get new content through our preference center, or unsubscribe. You can also remove yourself from all Tech in Asia emails over here. Copyright © 2026 Tech in Asia, All rights reserved. Tech in Asia (Podium Level 3), 1000 Toa Payoh North, News Centre, Singapore 318994 A member of The Business Times.