What events in the field of AI took place during the week

This week, the largest technology companies simultaneously placed bets on autonomous agents.: Microsoft is building an enterprise equivalent of the most popular open-source AI assistant, OpenAI and Anthropic have launched a race in cybersecurity, and NVIDIA has combined artificial intelligence (AI) with quantum computers. All the details are in the Izvestia article.

Microsoft is building an agent that never shuts down

Microsoft is testing the integration of features similar to the OpenClaw agent into Microsoft 365 Copilot. The new tool is aimed at corporate clients and, unlike the open OpenClaw, is created with enhanced security and management mechanisms.

According to sources, the key feature of the agent is constant background work without user involvement: it must independently perform multi—step tasks for a long time. It is still unclear whether the tool will work locally on the user's device or in the cloud — this issue remains crucial. The local launch on the Mac Mini has become a hallmark of the original OpenClaw: Apple's small computers have skyrocketed in sales in the wake of the agent's popularity.

A Microsoft official confirmed that the company is "experimenting" towards "orchestration and autonomy" for enterprise and consumer AI products. The new agent is scheduled to be shown at the Build conference in June.

Microsoft has already taken steps in this direction: in March, the company announced Copilot Cowork, an agent running inside Microsoft 365 applications based on its own Work IQ technology and the Anthropic Claude model, and in February it released Copilot Tasks in previews, focused on a wider range of tasks. At the same time, Claude remains the most popular model among OpenClaw users.

Goals and objectives: agencies have established standards for the implementation of neural networks

Why has the bar been set low for business and how will efficiency be determined?

OpenAI and Anthropic launched a cybersecurity model race

On April 14, OpenAI launched a limited rollout of GPT-5.4, a Cyber-specialized software vulnerability detection model. The release took place exactly one week after Anthropic restricted access to its own Mythos Preview model, capable of detecting thousands of vulnerabilities in the OS and popular open-source projects.

GPT-5.4-Cyber is available to participants of the Trusted Access for Cyber program, launched in February for verified security professionals. OpenAI has removed a number of standard sensing restrictions for this model. Fuad Mateen, a researcher at the company, called cyber defense a "team sport" and said that "no one should choose winners and losers" in the market for such tools. The two competitors follow opposite strategies: Anthropic has limited itself to about 40 partners within the closed Glasswing project, while OpenAI plans to gradually expand access to thousands of specialists.

One announcement by Anthropic brought down the shares of Adobe, Figma and Wix

The Information, citing a source familiar with the company's plans, announced preparations for the release of Claude Opus 4.7 and a new AI tool that creates websites and presentations from text queries. The launch is expected before the end of this week.

The news immediately affected the market: shares of Adobe, Figma, GoDaddy and Wix fell by 2-4%. Analysts regard the design tool of Anthropic as the company's direct entry into the market of website designers and presentations with a volume of about $30 billion. In addition, The Decoder reported that venture capital investors are already ready to value Anthropic at $800 billion, more than double the February estimate of $380 billion. The company's annual revenue grew from $9 billion to $30 billion in three months, and more than a thousand corporate clients now spend over $1 million a year each. At the same time, Anthropic switched to a new pricing model for Enterprise: instead of a fixed tariff of up to $200 per user, the company introduces a base fee of $20 plus payment for the consumption of computing resources.

The Code of the Universe: How AI searches for life outside the Solar System

Neural networks help astronomers analyze huge amounts of data in search of objects similar to Earth.

Google has taught Gemini to memorize your tasks in the browser

Google has announced the Skills feature in Chrome, a tool for saving Gemini promptos and launching them again with a single click on any pages and tabs at the same time. The saved script can be activated by entering the "/" symbol or by pressing the "+" button in the Gemini sidebar.

In addition to user scenarios, Google will provide a ready—made template library, in particular, for comparing product characteristics on several tabs at once or analyzing the composition of products. Templates can be edited and added to your personal collection.

NVIDIA combines AI with quantum computers

NVIDIA has introduced Ising: the world's first family of open-source AI models designed specifically to work with quantum computers. The models solve two key problems hindering the development of quantum systems: processor calibration (Ising Calibration) and real-time error correction of qubits (Ising Decoding).

Ising Decoding is 2.5 times faster and three times more accurate than existing industry error correction standards. The models are already being used at Harvard, Cornell University, IonQ and IQM Quantum Computers, as well as at Sandia National Laboratories. All models of the family are available on GitHub and Hugging Face with an open license.

All important news is on the Izvestia channel in the MAX messenger.