Concise AI news on new models, tools, companies, research findings and important developments shaping the industry. Each article explains what happened, why the news matters and what it can mean in practice for users, businesses and society.
According to Computerwoche, the right distribution of tasks between fixed rules, language models and autonomous agents decides whether AI pays off in a company. The article warns that uncontrolled token consumption can endanger the ROI and advises continuous review after the rollout.
In his post dated 28 October 2026, Simon Willison sees an inflection point in AI-supported game development. He describes several games that he says he created with Claude Opus 5.5 (as of October 2026) and brief instructions. This is the author’s assessment, not an independent review.
Gartner analyst Annette Zimmermann describes how companies are testing AI agents for administrative tasks and more complex processes. She also names obstacles such as costs, access management and regulatory limits in Europe.
This week, machines produce more than can currently be checked. One model rewrites the weights of another, an agent system reports 948 kernel vulnerabilities, and vendors report work that takes hours instead of weeks …
Google is permitting the use of external AI coding agents in Android Studio. According to the company, terminal tools are meant to speed up UI tasks and save tokens. Google is also presenting interfaces for local agent functions on mobile devices.
According to Surge AI, the best model tested reached a pass rate of only 10.8% on 50 realistic tax assignments. The yardstick was a 95% rubric.
With Claude Duet, Anthropic offers a proactive agent that, according to The Decoder, is meant to continue tasks on a virtual machine even when devices are switched off. As of October 2026, access initially runs through a waiting list for subscribers to the Pro and Max plans in the USA.
The Swedish company Tandem Health offers an AI assistant that documents doctors’ consultations. According to Handelsblatt, the firm wants to give healthcare professionals a billion hours back in total. Deutsche Telekom is taking a strategic stake.
According to CHIP, the real-time translation in Google Meet supports 63 languages in total and recognizes the spoken language by itself. Participants are meant to be able to display subtitles in their target language. With strong accents or people talking over one another, the quality can suffer.
Whether a model finds security vulnerabilities, tops a leaderboard or triggers orders: this week, the model’s own performance hardly decided how useful it was. What mattered was what is organized around it, …
According to the project description, OpenShell is a runtime environment that operates autonomous AI agents in isolated sandboxes. Policies are meant to control which files, system calls and network destinations the agents may access.
The FAZ describes a strongly growing market for AI solutions in legal advice. Law firms should choose tools according to their needs and keep attorney-client privilege, data privacy and the risk of invented statements in view.
According to Computer Bild, Google is trialling a Gemini function called “Call for Me” that makes phone calls for users, for instance to reserve a table. On this account it is so far intended only for owners of the Pixel 11 in the USA; a later extension is possible but not confirmed.
According to Computerwoche, DE-CIX is providing an MCP server through which network connections are meant to be controlled by natural language with AI agents and chatbots. The server is available as open-source software.
According to the Heise report, ChatGPT Voice can access external services such as email, calendar and Slack by voice. Security-relevant actions have to be confirmed on the screen.
Across this week, generating has become faster than checking. One model finds flaws in the core of PHP, but it takes months until these turn into installed updates, and that depends on server operators. …
Brain-computer interfaces are meant to read out neuronal signals and make them usable. According to the account of Futrue CEO Clemens Fischer, they could help paralyzed and blind people. Anonymized data is to flow into research for this, which raises data protection questions.
According to t3n, Google is working on Playground, a platform with which games are meant to be created and adjusted by text input – even without programming knowledge.
With the Voxtral 3 model family, Mistral AI aims to enable voice agents that listen, respond and use tools in real time. According to The Decoder, operation on an organisation’s own infrastructure or through EU-hosted APIs is intended.
Microsoft and Nvidia have presented a notebook that, according to the manufacturers, is meant to run AI computations directly on the device, even without an internet connection. According to Handelsblatt the first devices are planned for the autumn, in the premium price segment.
According to Computerwoche’s account, Barmer is relying on an AI department of its own to automate processes such as family insurance or the reimbursement of vaccination costs and to relieve staff. The information comes from the report and has not been independently verified.
AI finally leaves the lab this week. BMW plans savings in the three-digit million range through autonomous purchasing agents, robots fold napkins in restaurants, and language models solve mathematical …
The device uses the RTX Spark platform with a Grace CPU and Blackwell GPU for local AI applications. The laptop starts at 2,899 euros and is aimed at professional users as well as developers.
The US financial services provider Equifax uses artificial intelligence to fend off threats and has raised automation in the Security Operations Center to 50 percent. The company reports an increase in attack volume of around 30 percent.
Researchers have developed a technique that keeps AI agents from stagnating during autonomous problem solving. The method resets the conversation history while the workspace is retained.
As the first US state, Utah is allowing the use of artificial intelligence for diagnosis and the prescribing of medication in a pilot programme – limited at first to the treatment of acne.
Anthropic has extended its managed agents platform with the orchestration of several agents. A main agent can distribute tasks in real time to up to 1,000 sub-agents, of which 64 can work in parallel.
AI agents are leaving the lab and taking on real responsibility – as a store manager in San Francisco, as a development tool for OpenAI researchers, or as autonomous diagnostic software in Utah. This week it becomes clear that …
The company Vornac uses an AI agent for penetration tests that, unlike traditional methods, is meant to identify security gaps continuously. The system uses a model of its own and can be operated both in the cloud and locally.
A journalist tested an AI agent with access to his bank account and emails for two weeks. The system made calls, cancelled duplicate subscriptions and organised appointments – albeit with technical hurdles.
A study by Cisco and Omdia shows that more than half of IT managers already use agent-based AI tools in network operations. 99 per cent of respondents, however, demand strict safety measures before they trust the technology.
The SigenMate 2700 Ultra combines solar storage with AI-controlled energy management. The modular system offers 2.688 kWh of capacity in its basic version and can be expanded to more than 56 kWh.
At DevDay, OpenAI presented cloud-based agents that can take on tasks continuously. The systems, called Dots, are meant to automate routine tasks but require clear rules for access and responsibility.
This week marks a turning point: AI systems are stopping waiting and starting to act. With Dots, OpenAI is putting agents into the race that run permanently and work proactively – while at the same time the training …
A framework from Stanford turns research papers into digital agents that can be questioned. They can carry out analyses, validate results and communicate with one another.
OpenAI has presented Dots, a new kind of AI agent that stays permanently active on the basis of GPT-6 Astra and has a cloud computer of its own. Users in Europe, however, can only use the function to a limited extent.
GMKtec is bringing a compact computer to market that is meant to run large language models locally. The device has 192 GB of working memory and a Ryzen processor with integrated graphics.
Specialised laptops with powerful processors let companies run AI tasks directly on the device. This is meant to reduce the costs of cloud-based token billing.
Google is testing a function that lets the AI assistant Gemini conduct phone calls with businesses on its own. The feature is running initially as a beta version exclusively for certain Pixel users in the USA.
This week reveals a pattern that runs through the whole AI industry: the technology is faster than its application. While OpenAI, Google and Meta are presenting agents meant to make phone calls, do the shopping and …
Conversational systems based on machine learning influence opinions more effectively than human conversation partners. Research shows measurable changes in political views and belief in conspiracy theories through data-based argument.
The ChatGPT Voice assistant makes it possible to manage emails, calendar appointments and Slack messages by voice command. Users can delegate tasks without having to intervene manually.
A company employs only nine people but more than 100 AI agents – some of them in leadership roles. The founder deliberately keeps the headcount low in order to achieve efficiency through automation rather than through new hires.
A new language model called Bonsai 27B has been compressed by quantization techniques to the point where it can run on consumer hardware such as smartphones and laptops with less than 4 GB of memory – without a cloud connection.
More than half of German companies rely on artificial intelligence, yet only 11 percent let their AI agents work independently. Trust in autonomous systems has dropped considerably within a single year.
AI agents are independently taking on tasks that seemed unthinkable not long ago – from penetration tests to drone control to autonomous purchasing. Yet alongside these breakthroughs, security incidents are piling up …
Artificial intelligence can support security departments through automated monitoring, the consolidation of SOC processes and the continuous assessment of the security situation. Machine learning helps to identify threats faster and to automate repetitive tasks.
Google is extending the Gemini app with the ability to present complex scientific topics as interactive 3D models and simulations directly in the chat window. The function is being rolled out step by step for Pro users.
Algorithms search through millions of molecules, identify target molecules and predict safety risks. Pharmaceutical research is increasingly relying on computer-assisted methods to shorten development times and raise success rates.
In the transition from classic robotic process automation to AI-supported automation, many companies are repeating old mistakes: missing process selection, inadequate data quality and insufficient change management put automation projects at risk.