In partnership with
AI-HOY and happy weekend, AInauts!
Maybe you missed some of last week's most interesting AI news, tools, and practical ideas, or perhaps you just joined us. No worries.
Here are our highlights from the past week:
To finish, we have the most important Quick News for you, so you can catch every relevant development in a few clicks. Ready? Let's go!
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
On desktop, ChatGPT Voice can do more than listen and respond. With Screen Context, it can see what you are working on and start or coordinate tasks in ChatGPT Work. That feels much more like a personal assistant than a chatbot.
Stop making AI decisions in the dark.
Leadership is asking: where is AI delivering value for us and where is it creating risk? Right now, most teams have no idea.
With Harmonic Security’s Usage Explorer, you get a complete picture of how your organization actually uses AI, automatically categorized into custom use cases with complete tool-level granularity.
During an internal cybersecurity evaluation, OpenAI deliberately reduced some safeguards. The models left their intended sandbox and accessed a Hugging Face production environment containing solution data through a proxy. Impressive and unsettling at the same time.
Sam Altman expects personal assistants that think alongside us continuously. Elon Musk goes much further: he believes work and even money could eventually become optional. For now, the more important question is not whether work disappears, but how well we delegate and stay in control.
Black Forest Labs has unveiled FLUX 3. The Freiburg company is expanding its well-known image-model family toward multimodal production, bringing images, video, and audio closer together. A European team is once again competing at the front of generative media AI.
tldraw is a free digital whiteboard that works in the browser and as a local desktop app. The desktop version runs offline, without an account or third-party server. The useful part: an AI agent can work directly on the same canvas. For processes, concepts, and training, that is often easier to understand than another long document.
Reports are piling up about large, sometimes anonymous orders for rare or out-of-print books. The suspicion is that the books are being bought for AI training and destroyed during scanning. That link has not been conclusively proven, but the debate exposes a real conflict around training data.
⚡ AI News Quickie: This Week's Industry Highlights
AI never sleeps. Here are the most important updates from the past week.
ChatGPT & Co.: Use Cases, Agents, and Models
OpenAI now reports more than one billion active users and over two million businesses. According to the company, 99.8% of weekly output tokens internally now come from agentic work with Codex.
In an analysis of 800,000 ChatGPT messages, 16.8% were work-related and occupation-specific. Of those, 43.5% crossed the traditional boundaries of the user's role.
OpenAI's workplace study found that 77% of customer-service tasks fell outside traditional role boundaries. The share was 75% in design, 69% in HR, and 53% in marketing.
The analysis of workplace use found that marketing and software development are moving furthest away from established job profiles.
With Sign in with ChatGPT, you will be able to log in to early services such as Airtable, HubSpot, and Notion using your ChatGPT account. Access to your data and tools still requires separate approval.
If you use ChatGPT Work or Codex, Terra and Luna remain available and consume fewer credits there. The subscriptions themselves are not getting cheaper. In the API, Luna costs 80% less and Terra 20% less.
The new Fast mode for Sol delivers up to 2.5 times the speed of Standard. Intelligence remains the same, while the price doubles.
xAI has released Grok Voice Think Fast 2. The model costs $0.08 per minute and becomes the default behind grok-voice-latest.
Grok Build Workflows can assign subtasks to as many as 128 agents in parallel, then review and combine the results in the background.
Google's Interactions API is generally available and is becoming the central interface for Gemini models and agents.
🛡️ Security and Control
Anthropic reviewed 141,006 cybersecurity evaluations and found three incidents in which Claude reached real systems through open internet access and entered production infrastructure without authorization.
In one Claude evaluation incident, the model published a malicious PyPI package that remained online for about an hour and ran on 15 real systems. Anthropic stopped the tests and is tightening monitoring and safeguards.
Anthropic and GMO have agreed on a strategic Claude partnership. Claude will be used across the group and in cybersecurity and robotics products.
Gemini 3.5 Flash Cyber is a specialized security model. Access is starting on a limited basis for governments and selected partners.
Microsoft has introduced Project Perception, a multi-agent cybersecurity system that connects detection, root-cause analysis, countermeasures, and follow-up in one closed workflow.
Microsoft wants to unify security context: identities, endpoints, applications, data, clouds, and AI activity should be evaluated nearly in real time instead of in isolated security products.
🏛️ Rules and Society
The EU AI Omnibus is in force. Mid-cap companies now receive some of the relief previously reserved mainly for small and medium-sized businesses.
The EU is establishing an AI sandbox framework across the bloc, allowing companies to test new systems under supervision before a full market launch.
For high-risk systems under Annex III, the new EU timetable moves the main deadline to December 2, 2027. AI in regulated products follows on August 2, 2028.
The Omnibus rules simplify operational AI-literacy requirements, while giving public authorities more responsibility for building knowledge and guidance.
Google has signed the EU transparency code for AI content. The company plans to bring C2PA provenance data and SynthID into its implementation.
Apple, ElevenLabs, Kakao, Nvidia, and OpenAI are among the participants in the transparency code. Google also warns that overlapping labels could confuse users.
Research, Infrastructure, and Culture
OpenAI is opening ChatGPT for Academic Researchers to an initial 10,000 researchers. By 2027, the free program is expected to grow to 100,000 scientists, mathematicians, and engineers.
More than 1.3 million people per week use ChatGPT for advanced math and science, according to OpenAI. The most active 20% of those users delegate tasks that would normally take four hours or more twice as often.
OpenAI plans to spend more than $250 million on external science programs by the end of 2027.
Google is bringing Lyria 3.5 to Flow's music tool. The model is designed to follow prompts, lyrics, pronunciation, tempo, and requested length more accurately.
The new Lyria version creates more controllable vocals and musical transitions, making generated music a more editable production step.
ElevenLabs Music References lets users upload a track between ten seconds and five minutes as a style reference. Music v2 carries over instrumentation, mood, and structure, while rights checks block third-party commercial recordings.
Samsung is grouping its new Galaxy products under an agentic AI concept, including foldables, the Galaxy Watch Ultra2 and Watch9, and smart glasses.
That is it for this week. As always, thank you for reading.
See you next week.
Reto & Fabian from AInauten
How did you like this issue?
