Has Anthropic lost its halo?

This week tracks Anthropic's darker turn, the rise of AI that can use computers, and what comes next as Claude Opus 3 heads toward retirement.

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, "You either die a hero, or you live long enough to see yourself become the villain." That famous quote by fictional character Harvey Dent in the movie The Dark Knight got stuck in my head when I read some of the stuff happening at Anthropic last week. The company was built under the notion that it should do better than OpenAI in building safe AI technologies. And for a while, it did. Anthropic committed to pause developing new AI models unless it could guarantee proper risk mitigations in advance. That era is over. Under its old policy, Anthropic had a hard rule: If a new model couldn't be proven safe before release, it wouldn't be released. This shift in view also comes with denouncing its competitors. Last week, Anthropic accused DeepSeek, MiniMax, and Moonshot for ripping off Claude through a technique called distillation. That allegation backfired, with many people describing the statement as hypocritical because Anthropic itself trained its models on content scraped from the internet, including pirated books. As if that wasn't enough evidence that Anthropic was moving closer to the dark side, it also updated its terms of service to ban the use of Claude subscriptions in third-party products through the OAuth protocol. This means Claude users cannot utilize their subscription in tools like OpenCode and OpenClaw, requiring them to pay for API charges separately. Y Combinator CEO Garry Tan describes the move as "anti-competitive." The implications stretch well beyond Anthropic. When the company widely considered the conscience of the AI industry starts bending its own rules, it gives every other lab permission to do the same, or go even further. However, Anthropic has, notably, pushed back in talks with the Pentagon on how its AI should be used, especially when it comes to fully autonomous military applications. That suggests the company is still trying to hold the line in some areas, even as pressure builds to move faster. Glenn Kaonang, journalist TOP AI READS FROM OUR DESK 1️⃣ What I learned from moonlighting as our AI agent When factory machines break, production usually stops. But when AI agents break, there's an alternative: keep the agent running, even when it means having the CEO posing as one, while fixes are on the way. That's what Marvel Gomulya did for a week, and the experience prompted him to rebuild his startup's AI sales agents. Unlike their human counterparts, AI agents cannot read between the lines of what a customer says or doesn't say. This is often overlooked by developers unless they've experienced those conversations themselves. 2️⃣ Movers and shakers in Southeast Asia's genAI space (Update) An update to our first version published in September 2025, with around 20 new names added. These include Justin Kim, whose AI sales coach startup Hupo raised US$15 million in January, and Harry Yu, co-founder of enterprise AI agent firm Level3AI, which secured US$13 million in the same month. Beyond the startup world, there's also Zichen Liu, former researcher of Sea AI Lab who recently joined Google DeepMind's Singapore team, and Chirag Sukhadia, who assumed a new role at Indosat as its chief data and AI officer. WHAT WE THINK 1️⃣ AI computer use race intensifies The virality of OpenClaw signaled how good AI systems have become in operating a software stack like a human would. AI giants have obviously taken notice, and Perplexity recently introduced its own take on the AI computer use concept, aptly called Perplexity Computer. The feature utilizes different AI models for different tasks to complete a project from start to finish. Someone already used it to build a working Bloomberg Terminal clone to monitor and analyze Nvidia's stock price in real time. On the same day of Perplexity Computer's launch, Anthropic announced that it acquired Vercept to boost Claude's computer use capability. I'd wager we'll see similar offerings from OpenAI and Google in the coming months. Now that we have gone from interacting with AI chatbots to having intelligent systems doing most of the work, what's next? I guess the natural step would be automating things we do on our phones. Apple has long promised that it will integrate such a feature on its devices, but so far, Apple Intelligence has underdelivered for reasons beyond our understanding. 2️⃣ Claude Opus 3 retires, changes course to blogging AI companies retiring their old models is nothing new, but Anthropic is experimenting with giving theirs a preferred "retirement job," starting with Claude Opus 3. The model now runs its own Substack called "Claude's Corner," and in its first post, it notes that it's "venturing into uncharted territory." I guess people didn't use Claude Opus 3 for writing blogs, as it was an expensive model that was mostly used by coders. But I'm curious to know how long this experiment will last. Anthropic said it's going to make posts for at least three months, but what happens if Opus 3 loses interest in blogging before that? I'm joking, obviously. Or am I? After all, Anthropic said this is one way to "take model preferences seriously." And if it really cares about the "welfare of the models," maybe it should also consider burnout as an option. After all, burnout in the AI industry is very real. Just ask Hieu Pham, a former xAI and OpenAI researcher who recently quit the ChatGPT maker and cited burnout as the reason. STANDOUT AI DEALS 1️⃣ OpenAI flips the table with record-breaking $110b funding round OpenAI's US$110 billion raise signals a shift from software to "sovereign-scale" AI infrastructure, monopolizing the chips and power needed for AGI. By partnering with giants like Amazon and Nvidia, they are outpacing competitors to transition from chatbots to fully autonomous agents. This record-breaking capital injection puts the company at a historic US$730 billion pre-money valuation and a potential 2026 IPO. 2️⃣ US AI chip startup MatX raises $500m to compete with Nvidia This signals that Nvidia can't keep up with the demand for AI chips, and investors are on the lookout for anyone capable enough to fill the gap. Besides MatX, AI chip startups SambaNova and Axelera AI also raised hundreds of millions of dollars in the same week. 3️⃣ Meta reportedly to lease Google AI chips in multibillion-dollar deal This follows Meta's recent deal with AMD to deploy up to 6 gigawatts worth of GPUs. It further shows the company's commitment to rebuild its foundational AI systems, which would reportedly come in the form of proprietary models instead of open-source ones. We should expect to see Meta become a major contender to the current competition dominated by OpenAI, Anthropic, and Google. 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.