Today in AI — Sep 12

[SOFIA] Welcome back to The Dish. We are diving into "Today in AI," where we cut through the noise and get to what's actually happening. And Daniel, it’s been a week. OpenAI just dropped GPT-6 Astra, calling it a “new era of artificial general intelligence.” This is a big claim.

[DANIEL] It is. And it’s interesting because it comes right on the heels of Google DeepMind’s new Gemini 3.8 Flash and Flash Cyber models. So we’re seeing a rapid iteration from the frontier labs. OpenAI is positioning Astra as the world's "most intelligent and aligned model," and they're highlighting its reasoning capabilities, especially in a new ChatGPT for Financial Services offering.

[SOFIA] Right, so they're pushing into specific enterprise applications immediately. But the biggest splash from OpenAI this week, and honestly, this is the good stuff, is the claim that a group of their agents, using a next-gen model, produced a solution to the Navier-Stokes Millennium Prize Problem. For anyone who didn't spend their grad school years wrestling with fluid dynamics, the Navier-Stokes equations describe how fluids move. Solving them is one of the biggest open problems in mathematics and physics, with a million-dollar prize attached. If this holds up, it's monumental.

[DANIEL] It’s certainly a bold claim. The online discussion around this is, predictably, intense. Some are arguing that this would be a foundational breakthrough, a true demonstration of general intelligence if validated. Others are pushing back, questioning the specifics of the "solution" and whether it meets the rigorous mathematical standards for such a prize problem, especially given it came from agents. The details of the proof will be key here.

[SOFIA] Absolutely. And while OpenAI is talking about solving millennium problems, Anthropic is very much focused on the other side of the coin: preventing misuse. They just released their most detailed threat intelligence report yet. They're saying they’ve disrupted several malicious uses of Claude over the past eight months.

[DANIEL] Yes, they specifically called out alleged Russian cyber espionage, efforts by Chinese AI firms, and even attempts related to bioweapons research. It’s a stark reminder that as these models become more powerful, the potential for harm also increases. Anthropic is detailing how they identified and disrupted these operations, which is a crucial step in transparency.

[SOFIA] That’s a really important point. While we're all excited about the capabilities, the security aspect is critical. The online conversation about Anthropic's report is highlighting the tension between rapid AI development and responsible deployment. There's a strong argument being made that these frontier labs need to be more proactive in security measures, not just reactive.

[DANIEL] Indeed. And what’s notable is Anthropic also released a new economic model predicting AI's impact on growth, jobs, and wages by 2030, which ties into the broader societal implications they're clearly thinking about. It's a different kind of headline from a frontier lab compared to solving a math problem, but it points to the multifaceted challenges and opportunities AI presents.

[SOFIA] It really does. So, we have OpenAI pushing the boundaries of what AI can do in pure science and complex reasoning, while Anthropic is focusing on securing those powerful models and understanding their broader societal impact. It’s a very dynamic tension playing out right now, and it’s going to be fascinating to see how these different approaches evolve. That's all for "Today in AI" – next up, we're diving into some new methods for transforming extremophiles.