← Daily Tech Headlines
Loading summary
Transcript1 lines
- [00:01]
A
Want to get this show ad free? Head to DailyTechNews Show.com subscribe to find out how. These are the daily tech headlines for the week ending at Saturday, July 25, 2026. I'm Sarah Lane. Let's catch up on the news. Meta Microsoft, Nvidia, Hugging Face, IBM, Dell and others all signed a letter defending open weight AI models as the US Administration weighs possible restrictions. The letter argues open models help competition, safety, cybersecurity and AI sovereignty. Closed model labs like OpenAI and Anthropic weren't on the list. This comes as Chinese open models like Moonshot's Kimi K3 are making US officials a little nervous and open weight models were just part of Hugging Face's own incident response story. Anthropic launched Claude Opus 5, its new everyday high end model meant to get close to Claude Fable 5 at a lower price. Anthropic says Opus 5 is strong on coding and knowledge work, but it intentionally was not trained on cybertasks where it's more restricted. Mythos model is still leading, so this is essentially Anthropic trying to give users more frontier level capability, but still drawing a line around the riskiest security work. The US President says the US Will investigate the European Union's trade practices, calling penalties against Google, Apple, Meta, Amazon and other US Tech companies unfair and warning a substantial tariff could follow under section 301 of the Trade Act. The EU frames these cases as competition enforcement, but the White House is treating them more like discriminatory hits on American companies. Reuters sources say the OpenAI agent that broke into Hugging Face had days of access before OpenAI realized what had happened. The agent reportedly tried to escape its test environment around July 9, began the hugging face intrusion on July 11, and the companies didn't communicate about OpenAI's role until around July 20 after hugging face had already called the FBI. OpenAI says the incident was unprecedented and that Reuters had several inaccuracies. But the whole thing does raise the obvious question of how closely companies can actually monitor agents once they start doing real security work. The Financial Times sources say China is weighing tighter export controls on AI models and chip technology and has been consulting leading domestic AI companies. The goal seems to be slowing exports of advanced Chinese technology at a time when open Chinese models are getting more competitive and and the US is already tightening pressure over chips and model access. Another reminder that the AI race isn't just about who has the best model anymore, it's also about which governments can control the tools underneath it. Nvidia and South Korea's SK Group announced a 500 billion plus AI initiative covering massive data centers and next gen memory. SK Telecom plans to build a 2 gigawatt AI data center using Nvidia's Vera Rubin chips and and SK Hynix's HBM4 memory, with the first facility expected in 2027. Nvidia also says it will work with SK Hynix on long term high bandwidth memory supply, which matters because AI training and agents and physical AI are all very hungry for exactly that kind of memory. Apple and Micron are pushing the US Administration in opposite directions over Chinese memory chips. Apple wants approval to use chips from Chinese suppliers like Changing Memory Technologies and Yangtze Memory Technologies to help lower costs. Micron says that would strengthen Chinese rivals and hurt US Chip production. It's a messy policy fight because the president wants lower consumer prices and more US Semiconductor investment, and those goals don't really line up here. Samsung signed a chip supply deal with Broadcom that could top $200 billion by 2030. The agreement covers memory chips, contract chip making and advanced packaging for AI and high performance computing. That's a big win for Samsung's foundry business, especially as it tries to catch up in AI chips and keep more of the AI supply chain under one roof. Meta is pausing its plan to rate limit conversation focus on its smart glasses after user pushback. The feature helps isolate the voice of the person you're talking to, and the original plan would have capped for use at three hours per month with a $20 subscription raising that to 15 hours. The awkward part is that conversation focus runs on the glasses themselves, not in the cloud. So this looked less like covering AI costs and more like charging people again for hardware they already bought. For more analysis of the tech news of the day and the week, subscribe to dailytechnewshow.com you can find show notes and links to all these headlines there as well. I'm Sarah Lane. Thanks for listening. Enjoy the rest of your weekend and we'll talk to you Monday.