CameraMatics raises up to €49 million for AI in transportation

Irish company CameraMatics has secured an investment of up to €49 million to expand its video analytics and fleet intelligence platform across Europe and the United States. The system uses AI to identify risks such as driver fatigue, dangerous incidents and problems around the vehicle, providing operators with information about fleet safety and performance. CameraMatics – Investment of up to €49 million

Why it matters: this is an example of AI moving directly into the physical infrastructure of the economy. Rather than being just a tool for documents or conversations, AI continuously analyzes what is happening on the road and can indirectly intervene through warnings and operational decisions. For commercial transportation, such systems could mean fewer accidents and lower costs.

French AI startup Uncovr turns surgical footage into medical documentation

French startup Uncovr is developing a platform that automatically analyzes video footage recorded during surgical procedures and turns it into medical documentation and codes required for billing. The company has raised approximately €6 million in a funding round led by Index Ventures, with participation from Seedcamp, Frst and other investors, and its technology is already being tested in hospitals in Europe and the United States. FrenchWeb – Uncovr raises €6 million for surgical AI

Why it matters: AI is not being used here to replace doctors, but to turn information produced during surgery into a useful form for medical documentation. This is an important direction for digital medicine because a significant share of healthcare workers’ time is spent not on medical procedures themselves, but on documentation and administrative tasks.

X discovers a network of around 200,000 fake accounts linked to China

X announced the removal of a network of around 200,000 inauthentic accounts, which the company links to China. Around 200 of these accounts were reportedly actively distributing messages about AI data centers in the United States, claiming that they are increasing energy costs and putting pressure on electricity grids. An independent analysis cited by Business Insider, however, found that many of the accounts had very little activity or no engagement at all, leaving questions about the operation’s actual impact.

Why it matters: AI is becoming the subject of a new form of information warfare. Data centers, energy consumption and the cost of AI development are already controversial issues in the United States, while the use of networks of fake accounts can make it difficult to distinguish genuine public concerns from a coordinated campaign originating abroad.

Anthropic changes the rules again for Claude Code usage

Anthropic has changed the schedule for usage limits on Claude Code, one of the most widely used AI tools for programming. The additional limits introduced in May will not simply disappear at the end of August, and the company has announced that the period of expanded usage will continue into September before another adjustment to the limits. Information about the new schedule shows how carefully computing resources need to be managed when programmers use AI agents for long periods and complex projects. Details on the changes to Claude Code limits

Why it matters: AI programming agents do not consume resources only when they answer a question. They can read files, run programs, analyze errors and repeat these operations dozens or even hundreds of times. As usage becomes more intensive, computing costs become a concrete limit on how much an agent can accomplish without human intervention.

Grok is starting to behave like an agent that can buy products

Grok Bot has moved beyond simply recommending or searching for products to testing actual commercial actions. An example presented on August 30 shows an agent capable of going through the stages of an online purchase and reaching the point of placing an order, including for a Tesla Model Y. AlexTech.ai – Grok becomes a shopping agent

Why it matters: the difference is simple but significant: a chatbot tells you what product to buy, while an agent can attempt to buy the product for you. If such systems become sufficiently safe, a large part of online commerce could shift from “search and buy” to “tell the AI what you need and approve the transaction.”

A $19.4 billion AI data center is facing controversy over its energy infrastructure

A data center in Vineland, New Jersey, being developed as part of a project associated with Microsoft and Nebius, has become the center of controversy over energy infrastructure and its impact on the local community. The project, associated with an investment of approximately $19.4 billion and 100,000 Nvidia GB300 chips, has been accused by local groups of operating up to 62 gas turbines without all the necessary permits and of building a 1.5-million-gallon tank for liquefied natural gas. Tom's Hardware – Microsoft-backed AI data center faces backlash

Why it matters: AI requires so much energy that it is beginning to change the way data centers themselves are built. When existing electricity infrastructure is not sufficient, developers may turn to their own power sources, and the AI debate becomes a debate about noise, pollution, permits and the impact on local communities.

A new algorithm can make AI models up to three times faster at long reasoning tasks

Researchers from Stanford and other institutions have introduced Prefix Sliding, a method that means AI models no longer have to retain and reread every step of a very long reasoning process. The system keeps the initial instructions and a window containing the most recent information, while discarding some of the intermediate tokens (units of text processed by the model). In the reported tests, the method made some models approximately three times faster without additional training. ArXiv – Prefix Sliding for efficient test-time scaling

Why it matters: models that “think” for longer to solve difficult problems increasingly consume memory and computing power. If AI can retain only the information that matters instead of constantly rereading the entire process, agents could work longer and tackle more difficult problems without requiring infrastructure to grow at the same rate.

UK mobile networks risk falling behind as AI-generated traffic grows

UK telecom operators warn that mobile networks are not prepared for the increase in data traffic associated with AI. VodafoneThree says most upgrades can be carried out using existing infrastructure, but more than half require administrative approvals, slowing down network expansion. The Guardian – UK risks falling behind in AI race without faster telecoms upgrades

Why it matters: AI is not developing only inside data centers. Users will increasingly send images, videos, data and voice commands to AI models, and all of that information has to travel through networks. If telecom infrastructure cannot keep pace, the speed and availability of AI services could become a problem even when the models themselves are powerful enough.

New malware targeted organizations in Romania, Spain and Turkey

Recorded Future researchers identified HOOKEDGE, Windows malware used in campaigns targeting government, diplomatic and defense-industry organizations in Romania, Spain and Turkey. The campaigns took place between late September 2025 and early April 2026 and used Microsoft Word documents containing macros to install the malware. Recorded Future attributes the activity, with moderate confidence, to the BlueDelta group, which overlaps with APT28, Fancy Bear and Forest Blizzard and is associated by researchers with Russia’s military intelligence service, the GRU. Recorded Future – BlueDelta Targets Defense and Diplomacy with HOOKEDGE

Why it matters: Romania appears in a cyber-espionage campaign targeting strategically important European organizations. Recorded Future does not publicly identify the Romanian institutions targeted and does not state that they were compromised, so the information should be understood as documented targeting activity, not proof that a specific Romanian institution was successfully breached.

At TechBBQ, Europe increasingly debated who controls AI

At the TechBBQ conference in Copenhagen, held from August 27–29, European investors, founders and technology executives repeatedly returned to the question of who controls AI. The discussion focused not only on the models Europe can build, but also on who owns the infrastructure, data and models on which European companies depend. TechCrunch – Europe's AI conversations at TechBBQ

Why it matters: for Europe, “AI sovereignty” is becoming a very concrete issue: how vulnerable is the economy if an American or Chinese provider changes the rules, prices or availability of a model? The discussion naturally connects with Aidapted and its focus on the development of European AI infrastructure and projects through which Europe is trying to build its own capabilities.

Polish startup Codaro is building an “auditor” for AI agents that write code

Wrocław-based startup Codaro is developing infrastructure that allows companies to use AI agents for programming while also checking and controlling what those agents do. The system checks AI agents’ actions and looks for errors, security problems, exposed passwords or API keys and other risks before changes reach the final product. Codaro – Control infrastructure for AI coding agents Nowosci.ai – Wrocław startup Codaro

Why it matters: the more freedom AI agents receive to modify software, the more important it becomes to have a system that checks every action before it reaches a real product. A new category of software is therefore emerging: tools that do not perform the agent’s work, but audit it, checking whether the agent has done the job correctly and safely.

Sources