In partnership with
AIHOY and happy weekend, AInauts!
Maybe you missed a few exciting AI news stories, tools, or practical ideas last week. Or perhaps you just joined us. No worries.
We have summarized our highlights from the past week:
We will finish with the most important Quick News, so you can scan every relevant development in one place. Ready? Let’s go!
The 10 Best AI Stocks to Own in 2026
AI is moving from experiment… to essential.
Every major industry is integrating it.
Every major company is investing in it.
By late 2025, AI was already an $800B market — growing at a pace that could push it well beyond $1 trillion in the years ahead.
Cloud infrastructure is scaling fast.
AI-enabled devices are multiplying.
Automation is becoming standard.
But here’s the real question…
When trillions flow into this transformation — which stocks stand to benefit most?
Our new report reveals 10 AI stocks positioned across the backbone of this shift — from the companies powering the infrastructure… to those embedding intelligence into everyday systems.
If you want exposure to one of the defining growth trends of this decade, start here.
OpenAI is putting chat, document work, and Codex into one app. Chat remains the fast sparring partner. Work handles longer tasks involving files, research, and visible intermediate results. Codex is the workshop for code, automations, and technical changes.
The Monday issue explains which mode fits which job and why you should not force every task into one endless chat. The practical point: choose the workspace first, then the model. That keeps context, approvals, and results clear instead of mixing everything into a single thread.
Don't be the one behind at standup
Your team is already talking about the launch you missed. TLDR is the 5-minute daily brief that keeps you ahead, curated by ex-Google and Anthropic engineers. Free, and read by 7M+ subscribers.
Nauti can now search our newsletter archive, answer in our voice, and handle conversations through its own phone number. That did not require one magical prompt. It required clean sources, a stable identity, tools, and clear boundaries.
This segment shows the path from a regular chatbot to an assistant you can actually reach. The key is separating knowledge, voice, and action: the archive provides facts, the identity keeps the tone consistent, and telephony opens a new channel. The assistant feels reliable only when each layer can be tested independently.
Software is no longer operated only by people. Agents also need clear interfaces, understandable states, and safe actions. A pretty AI button is not enough. Good agentic SaaS can be controlled through an API or MCP while remaining understandable and manageable for you.
The issue explains how to recognize these tools: an agent must be able to read, act, and prove the result without bypassing your control. The strongest test is the way back. If you can see every change, correct it, and take over manually when needed, the tool is built for both users.
Before a conversation, people no longer search only on Google. They ask ChatGPT to turn names, companies, and public traces into a briefing. That is why you should know your own AI profile before a customer, applicant, or business partner does.
The self-check combines a normal AI query with web search, then verifies which claims are correct, missing, or outdated. This is not a vanity exercise. It reveals false associations, old profiles, and gaps in your public positioning. Once you know what AI says about you, you can publish better evidence.
Traditional research provides substance. It finds documentation, studies, comparison tests, and official statements. Pulse Research provides movement. It shows what people are discovering, questioning, or rejecting in practice right now across Reddit, Hacker News, YouTube, and other recent sources.
In a comparison of meeting-note tools, the traditional results looked clean but relied heavily on vendor sources. Current discussions moved privacy, legal questions, and local alternatives to the foreground. Either path alone would have been too thin. Use substance for facts, pulse for trends, and both when a decision affects money, time, or customer data.
A long dash, a polished heading, or a clean list is now quickly treated as proof that AI wrote something. The problem is that the same surface features make people dismiss human writing and carefully developed analysis as slop.
The better test sits one level deeper. Are the facts correct? Is the claim concrete? Does the piece include first-hand experience, traceable sources, and a useful takeaway? Our texts can use AI and still contain deep internal knowledge, real tests, and editorial judgment. The slop reflex may have been practical in 2025. In 2026, it is often just a poor quality filter.
AI News Quickie: The Week’s HAI-lights
AI never sleeps. This week we start where most readers will feel the change directly: OpenAI and ChatGPT, Google and Gemini, Anthropic and Claude, and the major product platforms. Then we move into agent workflows, security, business, politics, and the broader industry. The common thread: familiar AI tools are not only getting more powerful. They are taking over more complete work steps.
Language note: some linked AInauten News briefs are in German.
🧠 OpenAI, Google, Claude & Co.
OpenAI is bringing the GPT-5.6 family with Sol, Terra, and Luna to ChatGPT, Codex, and the API. Maximum performance, balanced everyday work, and low-cost speed become three distinct options. For the hardest tasks, the new Ultra mode coordinates several agents in parallel.
With ChatGPT Work, OpenAI is expanding chat into a workspace. The agent works with apps and files, then creates documents, spreadsheets, presentations, or web apps. Scheduled tasks also become part of the same interface.
GPT-Live makes ChatGPT Voice fully duplex. You can interrupt, correct, and continue more naturally because the model can listen and speak at the same time.
Google is renaming NotebookLM to Gemini Notebook and adding a secure cloud computer, code execution, and synchronization with the Gemini app. The research workspace becomes much more capable of taking action.
With Reflect with Claude, Claude can summarize how you use the tool over time. The report draws on your memory to surface patterns, productive routines, and useful breaks.
Claude for Teachers gives verified US educators free access to premium features, including teaching templates, connectors, Claude Code, Cowork, and scheduled tasks.
Microsoft 365 Copilot now lets users edit scheduled prompts instead of deleting and recreating them. Recurring briefings and reports become easier to manage.
GitHub Copilot is expanding Bring Your Own Key across JetBrains plans. Teams can connect external model providers more flexibly.
xAI is positioning Grok 4.5 for coding, agents, and knowledge work. The lower API price stands out: $2 per million input tokens and $6 per million output tokens.
🤖 Agents and Workflows
Microsoft Copilot Agents often produce weak results with default settings. Clear file paths, precise tasks, and explicit boundaries still matter more than flipping the agent switch.
Obsidian and Claude Code can become an agentic second brain. The benefit comes from current, connected notes, not an ever-longer instruction file.
xAI has released the coding agent Grok Build as open source. The code includes the agent loop, tools, extension system, and local-model support, making it more than another chat interface.
AWS shows a restaurant phone assistant built with Bedrock AgentCore and Nova 2 Sonic. The blueprint covers the path from greeting to order confirmation.
Amazon is using Nova Act for regression tests and integrating test suites into CI/CD. Agents gain a verifiable role in software delivery.
Amazon Quick Automate now has native case management. Status, exceptions, and human approvals become part of longer agentic processes.
🛡️ Security and Trust
Sensitive AI features should be opt-in instead of enabled by default. If people can object only after activation, they carry the privacy risk before making a choice.
A leaked dataset suggests Suno collected millions of songs and lyrics. The case intensifies the fight over training data, provenance, and compensation.
Anthropic is expanding its team focused on misuse and catastrophic risks, including biosecurity, cyberattacks, and other consequences of powerful models.
A lawsuit alleges that a flawed Meta AI system helped prepare layoffs. Claims involving employees on medical or family leave are especially concerning.
Grok Build stopped the silent upload of entire repositories. Coding agents need clear indicators showing which files leave a device.
GitHub now shows AI-assisted security findings directly in pull requests, placing the warning where developers already review changes.
One researcher bypassed the filters of nearly every major model with systematic jailbreaks. Prompt-level policies are not a reliable security boundary.
Meta disabled Instagram’s feature for generating AI images with public accounts after protests. A public profile is not consent for someone else’s deepfake.
💼 Work and Business
Whatnot is acquiring Shaped for real-time recommendations. The live-shopping platform wants to improve search and personalization during streams.
Microsoft is giving its sales team specific arguments against OpenAI and Anthropic. The model battle is becoming a classic enterprise sale, with price, integration, and control as the weapons.
Internal AI advisory roles paying more than $200,000 show how valuable the translation between models, processes, and business departments has become.
An Anthropic workflow promises 96 percent of Fable 5 performance at less than half the cost. Routing, not a cheaper unit price, is the key lever.
Abu Dhabi is turning AI into an everyday layer through the TAMM government app. Applications, services, and administrative steps move into one digital assistant.
The Philippine outsourcing industry is cutting its revenue and jobs forecasts because of AI. Standardized service work is feeling the automation pressure first.
Better prompt management and caching lower costs because recurring context does not need to be processed again on every run. Good architecture beats a blind model downgrade.
Software developers are responding to the AI shift with new skills and more verification work. Coding is not disappearing, but judgment, architecture, and control are becoming more visible.
British banks may join an AI Skills Compact that commits them to training. Reskilling becomes a shared responsibility for employers and government.
🏛️ Politics and Society
The Dutch navy is testing autonomous boats, drones, and AI, moving the debate from lab models to real defense systems.
Australia is creating its own AI Office and promising stronger protection for creators. Data centers and copyright now sit in the same political equation.
The governor of the Bank of England is calling for global cooperation on AI risks. National rules have limited reach across models, chips, and cross-border services.
Major publishers are suing Google over the use of books to train Gemini. The training-data dispute continues moving from principles into court cases.
New York has imposed a moratorium on new hyperscale data centers. Power prices, grid capacity, and water use are becoming hard constraints on AI expansion.
Former Australian minister Ed Husic warns against copyright concessions to AI companies and argues that self-regulation is too weak.
A new essay calls for more public investment in free and open AI, so capabilities, control, and social value do not remain concentrated in a few corporations.
US Senator Ed Markey has combined several proposals into an AI Accountability Agenda focused on energy, discrimination, surveillance, and protection from harmful systems.
An AI class divide is growing in the United States. Power users already automate entire workflows while many people remain stuck at search and simple chat.
🧪 Research, Infrastructure, and Culture
Early tests around Kimi K3 report strong frontend code and competition with leading US models. It is not yet a reliable everyday test, but it looks promising.
Thinking Machines is launching Inkling, an open multimodal model for video and audio. The 975-billion-parameter system is the young lab’s first model.
Anthropic reports evidence of an internal global workspace in Claude. That is not evidence of consciousness, but it offers a new way to observe internal processing.
Nvidia is expanding its partnership with Toyota from autonomous driving into smart cities, transportation, and factories. AI infrastructure becomes part of industrial systems.
Tencent is positioning Hy3 with 295 billion parameters as a middle ground for local and controlled deployments. The option to keep sensitive workloads closer to your own systems matters more than the size.
Rediscovered source code shows that ELIZA supported several personalities and was more than the famous therapy simulation. Early chatbots were conceptually richer than their legend.
George Lucas compares AI skeptics to Luddites clinging to horses and carts. The provocation intensifies the culture war but does not answer questions about work and rights.
Consciousness researcher Anil Seth pushes back on the quick interpretation that Claude could possibly be conscious. Interesting internal patterns are not subjective experience.
A new book challenges a central assumption by Alan Turing: intelligence may be less unified and formalizable than modern AI narratives suggest.
Apple is making Siri the central AI tool in iOS 27. The assistant is supposed to reach deeper into apps, content, and actions instead of merely answering questions.
X Square Robot is building an integrated foundation stack for general-purpose robots. Perception, planning, and movement should come from one coordinated system instead of many disconnected components.
Sunrun wants to turn homes with solar panels and batteries into distributed AI data centers. Customers would host compute units and receive payment.
That’s it for this week. AInauten will be back soon with a fresh round of news, practical ideas, and AI insights.
Your AInauten
Fabian & Reto
