7 AI News Today: Murati, Alexa, Android Protection

Today, once again, things are getting exciting on the front line of practical AI: language models are becoming more natural in real time, platforms are embedding AI directly into their products, and regulation remains the mandatory program everyone likes to ignore until it gets expensive. On top of that, there are fresh market moves between Anthropic and OpenAI, plus new image AI for developers.

In short: if you want to know where AI is actually heading right now, these are the relevant signals today. Not just nice demos, but things that influence products, workflows, and budgets.

🎙️ Murati’s startup wants to make voice AI finally fluent

Mira Murati’s new startup is launching its first AI model and pursuing a very clear goal: voice AI should no longer feel like a slightly overenthusiastic Q&A machine, but respond more naturally, faster, and in a way that can be interrupted. The model processes audio, video, and text in parallel in 200-millisecond chunks. This kind of real-time interaction is exactly where many voice systems today still stumble. According to the report, the company wants to outperform OpenAI’s GPT-Realtime-2 and Google’s Gemini Live in interaction quality. For product teams, that’s exciting because it opens the door to new assistants, coaching apps, or support tools that really stay “in the conversation” instead of just delivering answers. And yes: if AI soon becomes more talkative than some people in meetings, that’s no longer a bug, but apparently the feature.
Source: The Decoder

📜 EU AI Act: the clock is ticking for companies

heise breaks down in a webinar what companies need to have completed by August 2026 under the EU AI Act. That sounds dry, but in practice it is highly relevant: anyone using AI in products, processes, or internal workflows now has to deal with obligations, risk classes, documentation, and responsibilities. Especially in European companies, the big mistake is often not “we’re not doing anything,” but rather “we’re already doing something with AI somewhere” — and only realizing too late that this turns into regulatory work. The key point: compliance is not just a legal issue, but a product, IT, and management task. So if you’re introducing AI in your company, you shouldn’t wait until 2026 to start looking for the AI Act folder. Otherwise, it’ll find you.
Source: heise

🚗 EV subsidies: money only with paperwork

Starting Tuesday, applications can be submitted for the government subsidy for electric cars, including retroactively from the start of the year. Up to 6,000 euros is available — which naturally raises the question of how many people will suddenly develop an enormous, spontaneous interest in electromobility. This news matters for AI not because it announces a big model, but because it shows how strongly subsidy programs influence purchasing decisions and platform data. For manufacturers, dealers, and mobility platforms, that means demand can shift quickly, and whoever spots these signals early can plan sales, consulting, and inventory management better. For end users, the practical advice is mundane but important: check deadlines, conditions, and subsidy logic carefully before the application disappears into the administrative swamp.
Source: heise

🛡️ Google brings real-time AI protection to Android

Google is expanding Android protection with new AI features against phone scams and harmful app behavior. The exciting part is not just “more security,” but the real-time component: the system is designed to detect suspicious patterns before damage occurs, rather than only reacting afterward. This is especially important in the mobile environment, where attacks often rely on social engineering, manipulated apps, or dubious calls — exactly where classic signature-based logic quickly reaches its limits. For developers, this also means platforms will increasingly intervene in the behavior of apps and communication. For users, it’s good news, even if with every new protection feature you should pay a little attention to how much context the system is reading along the way. After all, security is usually a trade-off — and not always one with crystal-clear fine print.
Source: heise

🖼️ Luma makes image AI more affordable for developers

Luma is bringing its image model Uni-1.1 to API and positioning itself directly alongside the big names in the market. With prices starting at $0.04 per image and 2048-pixel output, the offering is interesting for many teams that want to do more than just test image generation — they want to build it into real products. What’s especially interesting is the combination of web search, built-in reasoning, and up to nine reference images. This makes the API more flexible for use cases like marketing assets, product visualizations, or creative assistants with style guides. According to the Arena ranking, Luma sits in third place behind Google and OpenAI — not quite at the very top, but already in the league where things get serious. For developers, that means more choice, more price pressure, and more possibilities. For the market, it means the fight for cheap, good image AI is increasingly API-driven.
Source: The Decoder

📈 Anthropic overtakes OpenAI among B2B customers — for now

According to the Ramp AI Index, 34.4 percent of tracked US companies now use Anthropic, while OpenAI stands at 32.3 percent. This is less a spectacular breakthrough than a fairly telling market indicator: Anthropic seems to be gaining strong traction in B2B, especially with teams that want to use AI productively and in a controllable way. At the same time, the lead is fragile — exactly the kind of market lead that could look different again tomorrow. Still, the trend shows that competition is no longer defined only by model quality, but by trust, integration capability, and suitability for enterprises. For decision-makers, that means: the “best” provider is not automatically the most popular in sales, and vice versa. Welcome to the phase where AI market shares are as volatile as the vendors’ roadmaps.
Source: The Decoder

🛍️ Alexa moves directly into Amazon’s shop

Amazon is integrating Alexa Plus directly into Amazon.com, making the shopping assistant a fixed part of search. Strategically, that’s clever: instead of hiding an assistant as an extra feature somewhere in an app, it gets placed directly where purchase intent already exists. For users, that could mean better product search, guided purchase advice, and less keyword fiddling. For Amazon, it’s another step toward steering the shopping experience more strongly through AI — and thereby tying together data, recommendations, and conversion more tightly. For the market, it’s a good example of how generative AI is landing in e-commerce applications: not as a chat gimmick, but as a hub for search, advice, and purchase decisions. Practical. And of course, no coincidence.
Source: The Verge

🛠️ Tool tip of the day

If you want to test image AI for a product, marketing, or prototyping, take a look at Luma’s Uni-1.1 API. The combination of a low-cost entry point, reference images, and web search makes the tool interesting for quick experiments. For teams that don’t want to build their own model stack right away, this is a pragmatic path into image generation. #

---

Don’t want to miss any news? Subscribe to the newsletter