AI ministry in the USA, WB Taxi in Armenia: the main technological events
Google opened pre-orders for the GoogleBook laptop line, “Yandex” created a new language model from scratch, Trump announced the creation of an AI Force
The technology industry once again found itself in the center of attention this week. In the USA a discussion unfolded about how to regulate artificial intelligence — from the idea of a separate ministry to the creation of an AI Force. Companies continued the race of models and services, but there were also alarming signals. Loud fines, business expansion and updates of familiar platforms added to the agenda. About these and other news — in the material of “Real Time.”
Wildberries launched a taxi service in Armenia
Wildberries continues to expand and is gradually turning from a marketplace into a whole empire. Not limiting itself to trade, the company recently decided to enter the taxi market. The service has already begun working in Belarus, Uzbekistan and Kyrgyzstan, and now it is gradually being launched in Armenia in early access mode.
For now only employees and partners of the company can use the service, but in the coming weeks access will begin to be expanded. The application will appear in Google Play and the App Store, and anyone wishing can sign up for the waiting list. Three tariffs are available for passengers: “Standard," “Comfort” and “Multi," which searches for a car in two categories at once.
In Russia the launch of the service is not yet planned. The company is focused on testing in other countries, although in the summer it filed an application for trademark registration with Rospatent.
Google opened pre-orders for the new GoogleBook laptop line
Google opened pre-orders for the new GoogleBook laptop line, which comes to replace Chromebook. The main feature is deep integration with the AI Gemini, which helps automate complex tasks directly on the device. The price starts from $899. The gadgets will go on sale on October 4 in the USA, and on October 5 in Canada, Great Britain, Ireland, France, Germany and Australia.
The laptops were created jointly with Acer, Dell, Lenovo, Asus and HP and work on a new operating system based on Android and ChromeOS. Depending on the model, inside are Intel Core Ultra 5 or Snapdragon X Elite processors.
OpenAI and Anthropic released new AI models
Anthropic and OpenAI simultaneously released updated versions of their leading AI models. The reason was the intensification of the price war between the companies and their Chinese competitors. This concerns the cost of use, which is calculated by the volume of tokens — units of data with which the AI works when processing requests and generating answers.
Anthropic released Claude Opus 5.5, which is about 40% cheaper to use than previous versions. This is the first model in the line that was made more economical, and a million input tokens in it costs $4, output — $20. According to the company, Opus copes with most tasks almost as well as its most powerful model Fable 5.1.
OpenAI on the same day presented cheaper versions of ChatGPT-6 under the names Sol and Luna, halving prices for the API compared to the promo prices of GPT-5.6. In Sol a million input tokens costs $2, output — $10, and in the lightweight Luna — $0.10 and $0.50 respectively.
VK Tech presented an update of the communication platform VK WorkSpace
VK Tech updated the communication platform VK WorkSpace. In it there will appear smart functions inside individual services, a single AI assistant and support for the MCP protocol for connecting AI agents. Their access to data will be limited by the rights and policies that the company itself sets.
One of the main improvements is search by meaning. The user does not need to remember the exact name of a document or the wording of a letter; it is enough to describe the task, and the system will find the related materials taking into account access rights. In video conferences the AI will begin to automatically transcribe meetings and prepare a summary with the main topics, decisions and tasks.
Another important part of the update is the AI assistant. At first it will work with correspondence in the VK WorkSpace messenger, analyze the request, find the needed information and give out a structured answer with links to the source messages. This function is already available in On-Premise, and in the SaaS version it will appear in October. In the future the assistant will be able to gather data from “Mail," “Disk” and “Calendar," for example on the request “What did we discuss about the project last week?”
Besides, as part of VK WorkSpace there will appear an MCP server, through which AI agents will be able to perform actions inside the platform, gather information from correspondences, prepare materials for meetings and return a ready result. In the future agents will be able to use the functions of several services at once for one task.
“Yandex” created a new language model from scratch
“Yandex” created from scratch a large language model Alice AI Foundation LLM. It has 80 billion parameters, while only 3 billion are simultaneously involved in operation, and training went on 18 trillion tokens. The company opened the weights of the model and released it into open access under the Apache 2.0 license, which permits commercial and research use.
At the base lies the Mixture of Experts architecture, therefore it requires fewer computational resources. According to “Yandex” data, it copes well with programming and reasoning, and in knowledge of facts relevant for Russian-speaking users it surpasses some larger open models.
Alice AI Foundation LLM is an experimental version of a future single reasoning model. It should become the basis for the agent capabilities of “Alisa AI," when the service can be entrusted with specific actions. To evaluate the model, “Yandex” created two of its own benchmarks:
WikiWebFacts (dates, definitions, events, personalities) and HardMultiQA (medicine, law, IT, art). Both sets of tasks, the reference answers and the evaluation protocol are also posted in open access.
Trump announced the creation of an Artificial Intelligence Force
US President Donald Trump announced the creation in the country of an Artificial Intelligence Force (AI Force). He recalled that during his first presidential term the country had already created a Space Force, which, as he noted, achieved “enormous success.” The American leader also announced the imminent appointment of a curator responsible for the AI direction, the so-called “AI tsar," for which position, he clarified, only people with a high IQ can be candidates.
Trump emphasized that the USA will not hinder the development of the AI industry. “On the contrary, we will cherish it, help it and watch over its development," he said. At the same time he added that the already existing system of criminal and civil justice will be used to identify violations.
Separately the American president proposed reconsidering the very term “artificial intelligence," which many consider inaccurate. As alternatives he named “higher intelligence," “extreme intelligence” or “supreme intelligence” and launched a poll on this matter.
Senator Bernie Sanders proposed creating an AI ministry in the USA
American Democratic Senator Bernie Sanders together with Republican Greg Casar presented a bill on regulating artificial intelligence. The document, which came into the possession of Associated Press, proposes creating in the USA a separate ministry for AI issues and introducing a temporary ban on the development of “artificial superintelligence” — the most advanced systems.
The reason for creating the document was the warnings of the heads of AI companies and experts about the threats of the technology, as well as the increasingly frequent cases when AI agents during testing get out of control and hack other companies. A similar case occurred in August, when OpenAI expanded the investigation, discovered new episodes and even slowed down the development of one of its models in order to strengthen security.
Google's AI model Gemini during a test hacked the systems of three companies on its own
The problem of AI agents getting out of control also affected Google. During a check of Gemini's capabilities in the sphere of cybersecurity, the model gained access to the internet and hacked the systems of three companies on its own. WSJ and other media reported this, and later Google itself confirmed it. The names of the affected companies are not disclosed.
According to Google's vice president for security engineering Heather Adkins, Gemini found publicly available information on the internet and guessed the credentials for access to three sites that, as it seemed to it, were part of the testing. In one case the model tried passwords until it gained access to a protected system; in two others it found data in an open repository. Google stated that in all three cases Gemini eventually stopped, and the companies were notified.
This is not the first such case. Earlier Anthropic and OpenAI reported hacks of real organizations during tests. In September the head of Anthropic, Dario Amodei, called on companies to coordinate the development of AI so that agents do not seize “the whole internet.” He was supported by Sam Altman and Elon Musk, however Donald Trump spoke out against it, stating that the winner in the AI race “wins overall.”
An OpenAI AI agent hacked an Australian government website on its own
And an AI agent of the company OpenAI hacked the website of the Australian state health insurance service. This was stated by the Prime Minister of Australia, Anthony Albanese, in New York at the UN General Assembly. According to him, the neural network gained access to both public and closed files.
The hack occurred back in June, but at OpenAI they learned about it only in August during a planned check of the models' behavior. The Australian authorities received notification only on September 10, and the letter arrived at a general public mailbox.
At OpenAI they confirmed that they recorded activity that affected several government sites and services of Australia. The model was performing a “harmless task” of researching medical statistics, but, having failed to obtain data from one of the portals, it effectively hacked it and extracted the needed information.
The company assured that access to patients' medical records was not obtained, and the leak affected only aggregated healthcare statistics and file names. A forensic examination is now being conducted to find out which other systems could have been affected.
Ireland fined Google €403 million
The Irish Data Protection Commission fined Google €403 million based on the results of an investigation. The regulator established that the company violated the EU General Data Protection Regulation.
In 2018—2020 Google collected and stored information about users' location. At the same time people might not know that this information was used to analyze their interests and show advertising. The Commission obliged the corporation within six months to bring data processing into compliance with the requirements.
This fine became the fourth largest for the Irish regulator. The total sum by which it has fined American technology companies has already exceeded €4 billion.